From: Anna Schumaker <[email protected]>
It's possible for an NFS server to go down but come back up with a
different IP address. These patches provide a way for administrators to
handle this issue by providing a new IP address for xprt sockets to
connect to.
This is a first draft of the code, so any thoughts or suggestions would
be greatly appreciated!
Anna
Anna Schumaker (7):
net: Add a /sys/net directory to sysfs
sunrpc: Create a sunrpc directory under /sys/net/
sunrpc: Create a net/ subdirectory in the sunrpc sysfs
sunrpc: Create per-rpc_clnt sysfs kobjects
sunrpc: Create a per-rpc_clnt file for managing the IP address
sunrpc: Prepare xs_connect() for taking NULL tasks
sunrpc: Connect to a new IP address provided by the user
include/linux/sunrpc/clnt.h | 1 +
include/net/sock.h | 4 +
net/socket.c | 8 ++
net/sunrpc/Makefile | 2 +-
net/sunrpc/clnt.c | 5 ++
net/sunrpc/sunrpc_syms.c | 8 ++
net/sunrpc/sysfs.c | 160 ++++++++++++++++++++++++++++++++++++
net/sunrpc/sysfs.h | 22 +++++
net/sunrpc/xprtsock.c | 3 +-
9 files changed, 211 insertions(+), 2 deletions(-)
create mode 100644 net/sunrpc/sysfs.c
create mode 100644 net/sunrpc/sysfs.h
--
2.29.2
From: Anna Schumaker <[email protected]>
This is where we'll put per-rpc_client related files
Signed-off-by: Anna Schumaker <[email protected]>
---
net/sunrpc/Makefile | 2 +-
net/sunrpc/sunrpc_syms.c | 8 ++++++++
net/sunrpc/sysfs.c | 21 +++++++++++++++++++++
net/sunrpc/sysfs.h | 13 +++++++++++++
4 files changed, 43 insertions(+), 1 deletion(-)
create mode 100644 net/sunrpc/sysfs.c
create mode 100644 net/sunrpc/sysfs.h
diff --git a/net/sunrpc/Makefile b/net/sunrpc/Makefile
index 9488600451e8..1c8de397d6ad 100644
--- a/net/sunrpc/Makefile
+++ b/net/sunrpc/Makefile
@@ -12,7 +12,7 @@ sunrpc-y := clnt.o xprt.o socklib.o xprtsock.o sched.o \
auth.o auth_null.o auth_unix.o \
svc.o svcsock.o svcauth.o svcauth_unix.o \
addr.o rpcb_clnt.o timer.o xdr.o \
- sunrpc_syms.o cache.o rpc_pipe.o \
+ sunrpc_syms.o cache.o rpc_pipe.o sysfs.o \
svc_xprt.o \
xprtmultipath.o
sunrpc-$(CONFIG_SUNRPC_DEBUG) += debugfs.o
diff --git a/net/sunrpc/sunrpc_syms.c b/net/sunrpc/sunrpc_syms.c
index 236fadc4a439..3b57efc692ec 100644
--- a/net/sunrpc/sunrpc_syms.c
+++ b/net/sunrpc/sunrpc_syms.c
@@ -24,6 +24,7 @@
#include <linux/sunrpc/xprtsock.h>
#include "sunrpc.h"
+#include "sysfs.h"
#include "netns.h"
unsigned int sunrpc_net_id;
@@ -103,6 +104,10 @@ init_sunrpc(void)
if (err)
goto out4;
+ err = rpc_sysfs_init();
+ if (err)
+ goto out5;
+
sunrpc_debugfs_init();
#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
rpc_register_sysctl();
@@ -111,6 +116,8 @@ init_sunrpc(void)
init_socket_xprt(); /* clnt sock transport */
return 0;
+out5:
+ unregister_rpc_pipefs();
out4:
unregister_pernet_subsys(&sunrpc_net_ops);
out3:
@@ -124,6 +131,7 @@ init_sunrpc(void)
static void __exit
cleanup_sunrpc(void)
{
+ rpc_sysfs_exit();
rpc_cleanup_clids();
rpcauth_remove_module();
cleanup_socket_xprt();
diff --git a/net/sunrpc/sysfs.c b/net/sunrpc/sysfs.c
new file mode 100644
index 000000000000..efff6977095c
--- /dev/null
+++ b/net/sunrpc/sysfs.c
@@ -0,0 +1,21 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (c) 2020 Anna Schumaker <[email protected]>
+ */
+#include <net/sock.h>
+
+//struct kobject *rpc_client_kobj;
+static struct kset *rpc_client_kset;
+
+int rpc_sysfs_init(void)
+{
+ rpc_client_kset = kset_create_and_add("sunrpc", NULL, net_kobj);
+ if (!rpc_client_kset)
+ return -ENOMEM;
+ return 0;
+}
+
+void rpc_sysfs_exit(void)
+{
+ kset_unregister(rpc_client_kset);
+}
diff --git a/net/sunrpc/sysfs.h b/net/sunrpc/sysfs.h
new file mode 100644
index 000000000000..93c3cd220506
--- /dev/null
+++ b/net/sunrpc/sysfs.h
@@ -0,0 +1,13 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (c) 2020 Anna Schumaker <[email protected]>
+ */
+#ifndef __SUNRPC_SYSFS_H
+#define __SUNRPC_SYSFS_H
+
+extern struct kobject *rpc_client_kobj;
+
+extern int rpc_sysfs_init(void);
+extern void rpc_sysfs_exit(void);
+
+#endif
--
2.29.2
From: Anna Schumaker <[email protected]>
These will eventually have files placed under them for sysfs operations.
Signed-off-by: Anna Schumaker <[email protected]>
---
include/linux/sunrpc/clnt.h | 1 +
net/sunrpc/clnt.c | 5 +++
net/sunrpc/sysfs.c | 61 +++++++++++++++++++++++++++++++++++++
net/sunrpc/sysfs.h | 8 +++++
4 files changed, 75 insertions(+)
diff --git a/include/linux/sunrpc/clnt.h b/include/linux/sunrpc/clnt.h
index 02e7a5863d28..503653720e18 100644
--- a/include/linux/sunrpc/clnt.h
+++ b/include/linux/sunrpc/clnt.h
@@ -71,6 +71,7 @@ struct rpc_clnt {
#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
struct dentry *cl_debugfs; /* debugfs directory */
#endif
+ void *cl_sysfs; /* sysfs directory */
/* cl_work is only needed after cl_xpi is no longer used,
* and that are of similar size
*/
diff --git a/net/sunrpc/clnt.c b/net/sunrpc/clnt.c
index 612f0a641f4c..02905eae5c0a 100644
--- a/net/sunrpc/clnt.c
+++ b/net/sunrpc/clnt.c
@@ -41,6 +41,7 @@
#include <trace/events/sunrpc.h>
#include "sunrpc.h"
+#include "sysfs.h"
#include "netns.h"
#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
@@ -300,6 +301,7 @@ static int rpc_client_register(struct rpc_clnt *clnt,
int err;
rpc_clnt_debugfs_register(clnt);
+ rpc_netns_sysfs_setup(clnt, net);
pipefs_sb = rpc_get_sb_net(net);
if (pipefs_sb) {
@@ -327,6 +329,7 @@ static int rpc_client_register(struct rpc_clnt *clnt,
out:
if (pipefs_sb)
rpc_put_sb_net(net);
+ rpc_netns_sysfs_destroy(clnt);
rpc_clnt_debugfs_unregister(clnt);
return err;
}
@@ -733,6 +736,7 @@ int rpc_switch_client_transport(struct rpc_clnt *clnt,
rpc_unregister_client(clnt);
__rpc_clnt_remove_pipedir(clnt);
+ rpc_netns_sysfs_destroy(clnt);
rpc_clnt_debugfs_unregister(clnt);
/*
@@ -879,6 +883,7 @@ static void rpc_free_client_work(struct work_struct *work)
* so they cannot be called in rpciod, so they are handled separately
* here.
*/
+ rpc_netns_sysfs_destroy(clnt);
rpc_clnt_debugfs_unregister(clnt);
rpc_free_clid(clnt);
rpc_clnt_remove_pipedir(clnt);
diff --git a/net/sunrpc/sysfs.c b/net/sunrpc/sysfs.c
index afeaec79a9c7..dd298b9c13e8 100644
--- a/net/sunrpc/sysfs.c
+++ b/net/sunrpc/sysfs.c
@@ -2,7 +2,9 @@
/*
* Copyright (c) 2020 Anna Schumaker <[email protected]>
*/
+#include <linux/sunrpc/clnt.h>
#include <net/sock.h>
+#include "sysfs.h"
struct kobject *rpc_client_kobj;
static struct kset *rpc_client_kset;
@@ -53,8 +55,67 @@ int rpc_sysfs_init(void)
return 0;
}
+static void rpc_netns_client_release(struct kobject *kobj)
+{
+ struct rpc_netns_client *c;
+
+ c = container_of(kobj, struct rpc_netns_client, kobject);
+ kfree(c);
+}
+
+static const void *rpc_netns_client_namespace(struct kobject *kobj)
+{
+ return container_of(kobj, struct rpc_netns_client, kobject)->net;
+}
+
+static struct kobj_type rpc_netns_client_type = {
+ .release = rpc_netns_client_release,
+ .sysfs_ops = &kobj_sysfs_ops,
+ .namespace = rpc_netns_client_namespace,
+};
+
void rpc_sysfs_exit(void)
{
kobject_put(rpc_client_kobj);
kset_unregister(rpc_client_kset);
}
+
+static struct rpc_netns_client *rpc_netns_client_alloc(struct kobject *parent,
+ struct net *net, int clid)
+{
+ struct rpc_netns_client *p;
+
+ p = kzalloc(sizeof(*p), GFP_KERNEL);
+ if (p) {
+ p->net = net;
+ p->kobject.kset = rpc_client_kset;
+ if (kobject_init_and_add(&p->kobject, &rpc_netns_client_type,
+ parent, "%d", clid) == 0)
+ return p;
+ kobject_put(&p->kobject);
+ }
+ return NULL;
+}
+
+void rpc_netns_sysfs_setup(struct rpc_clnt *clnt, struct net *net)
+{
+ struct rpc_netns_client *rpc_client;
+
+ rpc_client = rpc_netns_client_alloc(rpc_client_kobj, net, clnt->cl_clid);
+ if (rpc_client) {
+ clnt->cl_sysfs = rpc_client;
+ kobject_uevent(&rpc_client->kobject, KOBJ_ADD);
+ }
+}
+
+void rpc_netns_sysfs_destroy(struct rpc_clnt *clnt)
+{
+ struct rpc_netns_client *rpc_client = clnt->cl_sysfs;
+
+ if (rpc_client) {
+ kobject_uevent(&rpc_client->kobject, KOBJ_REMOVE);
+ kobject_del(&rpc_client->kobject);
+ kobject_put(&rpc_client->kobject);
+ clnt->cl_sysfs = NULL;
+ }
+}
diff --git a/net/sunrpc/sysfs.h b/net/sunrpc/sysfs.h
index 93c3cd220506..279a836594e7 100644
--- a/net/sunrpc/sysfs.h
+++ b/net/sunrpc/sysfs.h
@@ -5,9 +5,17 @@
#ifndef __SUNRPC_SYSFS_H
#define __SUNRPC_SYSFS_H
+struct rpc_netns_client {
+ struct kobject kobject;
+ struct net *net;
+};
+
extern struct kobject *rpc_client_kobj;
extern int rpc_sysfs_init(void);
extern void rpc_sysfs_exit(void);
+void rpc_netns_sysfs_setup(struct rpc_clnt *clnt, struct net *net);
+void rpc_netns_sysfs_destroy(struct rpc_clnt *clnt);
+
#endif
--
2.29.2
From: Anna Schumaker <[email protected]>
And since we're using IP here, restrict to only creating sysfs files for
TCP and RDMA connections.
Signed-off-by: Anna Schumaker <[email protected]>
---
net/sunrpc/sysfs.c | 32 ++++++++++++++++++++++++++++++++
net/sunrpc/sysfs.h | 1 +
2 files changed, 33 insertions(+)
diff --git a/net/sunrpc/sysfs.c b/net/sunrpc/sysfs.c
index dd298b9c13e8..537d83635670 100644
--- a/net/sunrpc/sysfs.c
+++ b/net/sunrpc/sysfs.c
@@ -3,6 +3,7 @@
* Copyright (c) 2020 Anna Schumaker <[email protected]>
*/
#include <linux/sunrpc/clnt.h>
+#include <linux/sunrpc/addr.h>
#include <net/sock.h>
#include "sysfs.h"
@@ -55,6 +56,23 @@ int rpc_sysfs_init(void)
return 0;
}
+static ssize_t rpc_netns_address_show(struct kobject *kobj,
+ struct kobj_attribute *attr, char *buf)
+{
+ struct rpc_netns_client *c = container_of(kobj,
+ struct rpc_netns_client, kobject);
+ struct rpc_clnt *clnt = c->clnt;
+ struct rpc_xprt *xprt = rcu_dereference(clnt->cl_xprt);
+
+ return rpc_ntop((struct sockaddr *)&xprt->addr, buf, PAGE_SIZE);
+}
+
+static ssize_t rpc_netns_address_store(struct kobject *kobj,
+ struct kobj_attribute *attr, const char *buf, size_t count)
+{
+ return count;
+}
+
static void rpc_netns_client_release(struct kobject *kobj)
{
struct rpc_netns_client *c;
@@ -68,8 +86,17 @@ static const void *rpc_netns_client_namespace(struct kobject *kobj)
return container_of(kobj, struct rpc_netns_client, kobject)->net;
}
+static struct kobj_attribute rpc_netns_client_address = __ATTR(address,
+ 0644, rpc_netns_address_show, rpc_netns_address_store);
+
+static struct attribute *rpc_netns_client_attrs[] = {
+ &rpc_netns_client_address.attr,
+ NULL,
+};
+
static struct kobj_type rpc_netns_client_type = {
.release = rpc_netns_client_release,
+ .default_attrs = rpc_netns_client_attrs,
.sysfs_ops = &kobj_sysfs_ops,
.namespace = rpc_netns_client_namespace,
};
@@ -100,10 +127,15 @@ static struct rpc_netns_client *rpc_netns_client_alloc(struct kobject *parent,
void rpc_netns_sysfs_setup(struct rpc_clnt *clnt, struct net *net)
{
struct rpc_netns_client *rpc_client;
+ struct rpc_xprt *xprt = rcu_dereference(clnt->cl_xprt);
+
+ if (!(xprt->prot & (IPPROTO_TCP | XPRT_TRANSPORT_RDMA)))
+ return;
rpc_client = rpc_netns_client_alloc(rpc_client_kobj, net, clnt->cl_clid);
if (rpc_client) {
clnt->cl_sysfs = rpc_client;
+ rpc_client->clnt = clnt;
kobject_uevent(&rpc_client->kobject, KOBJ_ADD);
}
}
diff --git a/net/sunrpc/sysfs.h b/net/sunrpc/sysfs.h
index 279a836594e7..137a12c87954 100644
--- a/net/sunrpc/sysfs.h
+++ b/net/sunrpc/sysfs.h
@@ -8,6 +8,7 @@
struct rpc_netns_client {
struct kobject kobject;
struct net *net;
+ struct rpc_clnt *clnt;
};
extern struct kobject *rpc_client_kobj;
--
2.29.2
From: Anna Schumaker <[email protected]>
We won't have a task structure when we go to change IP addresses, so
check for one before calling the WARN_ON() to avoid crashing.
Signed-off-by: Anna Schumaker <[email protected]>
---
net/sunrpc/xprtsock.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/net/sunrpc/xprtsock.c b/net/sunrpc/xprtsock.c
index c56a66cdf4ac..250abf1aa018 100644
--- a/net/sunrpc/xprtsock.c
+++ b/net/sunrpc/xprtsock.c
@@ -2311,7 +2311,8 @@ static void xs_connect(struct rpc_xprt *xprt, struct rpc_task *task)
struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
unsigned long delay = 0;
- WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
+ if (task)
+ WARN_ON_ONCE(!xprt_lock_connect(xprt, task, transport));
if (transport->sock != NULL) {
dprintk("RPC: xs_connect delayed xprt %p for %lu "
--
2.29.2
From: Anna Schumaker <[email protected]>
We preserve the same port number, rather than providing a way to change
it. This keeps the implementation simpler for now.
Signed-off-by: Anna Schumaker <[email protected]>
---
net/sunrpc/sysfs.c | 7 +++++++
1 file changed, 7 insertions(+)
diff --git a/net/sunrpc/sysfs.c b/net/sunrpc/sysfs.c
index 537d83635670..47a7c9b8b143 100644
--- a/net/sunrpc/sysfs.c
+++ b/net/sunrpc/sysfs.c
@@ -70,6 +70,13 @@ static ssize_t rpc_netns_address_show(struct kobject *kobj,
static ssize_t rpc_netns_address_store(struct kobject *kobj,
struct kobj_attribute *attr, const char *buf, size_t count)
{
+ struct rpc_netns_client *c = container_of(kobj,
+ struct rpc_netns_client, kobject);
+ struct rpc_clnt *clnt = c->clnt;
+ struct rpc_xprt *xprt = rcu_dereference(clnt->cl_xprt);
+ struct sockaddr *saddr = (struct sockaddr *)&xprt->addr;
+
+ xprt->addrlen = rpc_pton(xprt->xprt_net, buf, count - 1, saddr, sizeof(*saddr));
return count;
}
--
2.29.2
From: Anna Schumaker <[email protected]>
Sunrpc will want to place its files under this directory
Signed-off-by: Anna Schumaker <[email protected]>
---
include/net/sock.h | 4 ++++
net/socket.c | 8 ++++++++
2 files changed, 12 insertions(+)
diff --git a/include/net/sock.h b/include/net/sock.h
index bdc4323ce53c..cd986c6fbc6f 100644
--- a/include/net/sock.h
+++ b/include/net/sock.h
@@ -1765,6 +1765,10 @@ void sk_common_release(struct sock *sk);
/* Initialise core socket variables */
void sock_init_data(struct socket *sock, struct sock *sk);
+
+/* /sys/net */
+extern struct kobject *net_kobj;
+
/*
* Socket reference counting postulates.
*
diff --git a/net/socket.c b/net/socket.c
index 33e8b6c4e1d3..43986419d673 100644
--- a/net/socket.c
+++ b/net/socket.c
@@ -141,6 +141,10 @@ static void sock_show_fdinfo(struct seq_file *m, struct file *f)
#define sock_show_fdinfo NULL
#endif
+/* /sys/net */
+struct kobject *net_kobj;
+EXPORT_SYMBOL_GPL(net_kobj);
+
/*
* Socket files have a set of 'special' operations as well as the generic file ones. These don't appear
* in the operation structures but are done directly via the socketcall() multiplexor.
@@ -3060,6 +3064,10 @@ static int __init sock_init(void)
goto out_mount;
}
+ net_kobj = kobject_create_and_add("net", NULL);
+ if (!net_kobj)
+ printk(KERN_WARNING "%s: kobj create error\n", __func__);
+
/* The real protocol initialization is performed in later initcalls.
*/
--
2.29.2
From: Anna Schumaker <[email protected]>
For network namespace separation.
Signed-off-by: Anna Schumaker <[email protected]>
---
net/sunrpc/sysfs.c | 41 ++++++++++++++++++++++++++++++++++++++++-
1 file changed, 40 insertions(+), 1 deletion(-)
diff --git a/net/sunrpc/sysfs.c b/net/sunrpc/sysfs.c
index efff6977095c..afeaec79a9c7 100644
--- a/net/sunrpc/sysfs.c
+++ b/net/sunrpc/sysfs.c
@@ -4,18 +4,57 @@
*/
#include <net/sock.h>
-//struct kobject *rpc_client_kobj;
+struct kobject *rpc_client_kobj;
static struct kset *rpc_client_kset;
+static void rpc_netns_object_release(struct kobject *kobj)
+{
+ kfree(kobj);
+}
+
+static const struct kobj_ns_type_operations *rpc_netns_object_child_ns_type(
+ struct kobject *kobj)
+{
+ return &net_ns_type_operations;
+}
+
+static struct kobj_type rpc_netns_object_type = {
+ .release = rpc_netns_object_release,
+ .sysfs_ops = &kobj_sysfs_ops,
+ .child_ns_type = rpc_netns_object_child_ns_type,
+};
+
+static struct kobject *rpc_netns_object_alloc(const char *name,
+ struct kset *kset, struct kobject *parent)
+{
+ struct kobject *kobj;
+ kobj = kzalloc(sizeof(*kobj), GFP_KERNEL);
+ if (kobj) {
+ kobj->kset = kset;
+ if (kobject_init_and_add(kobj, &rpc_netns_object_type,
+ parent, "%s", name) == 0)
+ return kobj;
+ kobject_put(kobj);
+ }
+ return NULL;
+}
+
int rpc_sysfs_init(void)
{
rpc_client_kset = kset_create_and_add("sunrpc", NULL, net_kobj);
if (!rpc_client_kset)
return -ENOMEM;
+ rpc_client_kobj = rpc_netns_object_alloc("net", rpc_client_kset, NULL);
+ if (!rpc_client_kobj) {
+ kset_unregister(rpc_client_kset);
+ rpc_client_kset = NULL;
+ return -ENOMEM;
+ }
return 0;
}
void rpc_sysfs_exit(void)
{
+ kobject_put(rpc_client_kobj);
kset_unregister(rpc_client_kset);
}
--
2.29.2
Hi Anna-
> On Jan 11, 2021, at 4:41 PM, [email protected] wrote:
>
> From: Anna Schumaker <[email protected]>
>
> It's possible for an NFS server to go down but come back up with a
> different IP address. These patches provide a way for administrators to
> handle this issue by providing a new IP address for xprt sockets to
> connect to.
>
> This is a first draft of the code, so any thoughts or suggestions would
> be greatly appreciated!
One implementation question, one future question.
Would /sys/kernel/net be a little better? or /sys/kernel/sunrpc ?
Do you have a plan to integrate support for fs_locations to probe
servers for alternate IP addresses? Would that be a userspace
utility that would plug values into this new /sys API?
> Anna
>
>
> Anna Schumaker (7):
> net: Add a /sys/net directory to sysfs
> sunrpc: Create a sunrpc directory under /sys/net/
> sunrpc: Create a net/ subdirectory in the sunrpc sysfs
> sunrpc: Create per-rpc_clnt sysfs kobjects
> sunrpc: Create a per-rpc_clnt file for managing the IP address
> sunrpc: Prepare xs_connect() for taking NULL tasks
> sunrpc: Connect to a new IP address provided by the user
>
> include/linux/sunrpc/clnt.h | 1 +
> include/net/sock.h | 4 +
> net/socket.c | 8 ++
> net/sunrpc/Makefile | 2 +-
> net/sunrpc/clnt.c | 5 ++
> net/sunrpc/sunrpc_syms.c | 8 ++
> net/sunrpc/sysfs.c | 160 ++++++++++++++++++++++++++++++++++++
> net/sunrpc/sysfs.h | 22 +++++
> net/sunrpc/xprtsock.c | 3 +-
> 9 files changed, 211 insertions(+), 2 deletions(-)
> create mode 100644 net/sunrpc/sysfs.c
> create mode 100644 net/sunrpc/sysfs.h
>
> --
> 2.29.2
>
--
Chuck Lever
On Tue, Jan 12, 2021 at 08:09:09AM -0500, Chuck Lever wrote:
> Hi Anna-
>
> > On Jan 11, 2021, at 4:41 PM, [email protected] wrote:
> >
> > From: Anna Schumaker <[email protected]>
> >
> > It's possible for an NFS server to go down but come back up with a
> > different IP address. These patches provide a way for administrators to
> > handle this issue by providing a new IP address for xprt sockets to
> > connect to.
> >
> > This is a first draft of the code, so any thoughts or suggestions would
> > be greatly appreciated!
>
> One implementation question, one future question.
>
> Would /sys/kernel/net be a little better? or /sys/kernel/sunrpc ?
>
> Do you have a plan to integrate support for fs_locations to probe
> servers for alternate IP addresses? Would that be a userspace
> utility that would plug values into this new /sys API?
We already have dns resolution for fs_locations, right? Why can't we
use that here? Is it that the mount call doesn't give us a host name?
Or we don't trust dns to have the updated IP address for some reason?
--b.
>
>
> > Anna
> >
> >
> > Anna Schumaker (7):
> > net: Add a /sys/net directory to sysfs
> > sunrpc: Create a sunrpc directory under /sys/net/
> > sunrpc: Create a net/ subdirectory in the sunrpc sysfs
> > sunrpc: Create per-rpc_clnt sysfs kobjects
> > sunrpc: Create a per-rpc_clnt file for managing the IP address
> > sunrpc: Prepare xs_connect() for taking NULL tasks
> > sunrpc: Connect to a new IP address provided by the user
> >
> > include/linux/sunrpc/clnt.h | 1 +
> > include/net/sock.h | 4 +
> > net/socket.c | 8 ++
> > net/sunrpc/Makefile | 2 +-
> > net/sunrpc/clnt.c | 5 ++
> > net/sunrpc/sunrpc_syms.c | 8 ++
> > net/sunrpc/sysfs.c | 160 ++++++++++++++++++++++++++++++++++++
> > net/sunrpc/sysfs.h | 22 +++++
> > net/sunrpc/xprtsock.c | 3 +-
> > 9 files changed, 211 insertions(+), 2 deletions(-)
> > create mode 100644 net/sunrpc/sysfs.c
> > create mode 100644 net/sunrpc/sysfs.h
> >
> > --
> > 2.29.2
> >
>
> --
> Chuck Lever
>
>
On Tue, Jan 12, 2021 at 11:59 AM J. Bruce Fields <[email protected]> wrote:
>
> On Tue, Jan 12, 2021 at 08:09:09AM -0500, Chuck Lever wrote:
> > Hi Anna-
> >
> > > On Jan 11, 2021, at 4:41 PM, [email protected] wrote:
> > >
> > > From: Anna Schumaker <[email protected]>
> > >
> > > It's possible for an NFS server to go down but come back up with a
> > > different IP address. These patches provide a way for administrators to
> > > handle this issue by providing a new IP address for xprt sockets to
> > > connect to.
> > >
> > > This is a first draft of the code, so any thoughts or suggestions would
> > > be greatly appreciated!
> >
> > One implementation question, one future question.
> >
> > Would /sys/kernel/net be a little better? or /sys/kernel/sunrpc ?
Possibly! I was trying to match /sys/fs/nfs, but I can definitely
change this if another location is better.
> >
> > Do you have a plan to integrate support for fs_locations to probe
> > servers for alternate IP addresses? Would that be a userspace
> > utility that would plug values into this new /sys API?
Yeah, I would expect there to be a new utility to help with assigning
new values. I haven't given any thought to using fs_locations yet, but
it could probably work.
>
> We already have dns resolution for fs_locations, right? Why can't we
> use that here? Is it that the mount call doesn't give us a host name?
> Or we don't trust dns to have the updated IP address for some reason?
The mount call doesn't give us a host name (that I can find, at
least). By the time we get to the sunrpc layer we're dealing with just
the IP address anyway. I'd expect there to be a userland utility to
translate the dns name to the new IP address and pass it along to the
new API.
Anna
>
> --b.
>
> >
> >
> > > Anna
> > >
> > >
> > > Anna Schumaker (7):
> > > net: Add a /sys/net directory to sysfs
> > > sunrpc: Create a sunrpc directory under /sys/net/
> > > sunrpc: Create a net/ subdirectory in the sunrpc sysfs
> > > sunrpc: Create per-rpc_clnt sysfs kobjects
> > > sunrpc: Create a per-rpc_clnt file for managing the IP address
> > > sunrpc: Prepare xs_connect() for taking NULL tasks
> > > sunrpc: Connect to a new IP address provided by the user
> > >
> > > include/linux/sunrpc/clnt.h | 1 +
> > > include/net/sock.h | 4 +
> > > net/socket.c | 8 ++
> > > net/sunrpc/Makefile | 2 +-
> > > net/sunrpc/clnt.c | 5 ++
> > > net/sunrpc/sunrpc_syms.c | 8 ++
> > > net/sunrpc/sysfs.c | 160 ++++++++++++++++++++++++++++++++++++
> > > net/sunrpc/sysfs.h | 22 +++++
> > > net/sunrpc/xprtsock.c | 3 +-
> > > 9 files changed, 211 insertions(+), 2 deletions(-)
> > > create mode 100644 net/sunrpc/sysfs.c
> > > create mode 100644 net/sunrpc/sysfs.h
> > >
> > > --
> > > 2.29.2
> > >
> >
> > --
> > Chuck Lever
> >
> >
> On Jan 13, 2021, at 2:23 PM, Anna Schumaker <[email protected]> wrote:
>
> On Tue, Jan 12, 2021 at 11:59 AM J. Bruce Fields <[email protected]> wrote:
>>
>> On Tue, Jan 12, 2021 at 08:09:09AM -0500, Chuck Lever wrote:
>>> Hi Anna-
>>>
>>>> On Jan 11, 2021, at 4:41 PM, [email protected] wrote:
>>>>
>>>> From: Anna Schumaker <[email protected]>
>>>>
>>>> It's possible for an NFS server to go down but come back up with a
>>>> different IP address. These patches provide a way for administrators to
>>>> handle this issue by providing a new IP address for xprt sockets to
>>>> connect to.
>>>>
>>>> This is a first draft of the code, so any thoughts or suggestions would
>>>> be greatly appreciated!
>>>
>>> One implementation question, one future question.
>>>
>>> Would /sys/kernel/net be a little better? or /sys/kernel/sunrpc ?
>
> Possibly! I was trying to match /sys/fs/nfs, but I can definitely
> change this if another location is better.
Ah... since this is a supplement to the mount() interface, maybe
placing this new API under /sys/fs/nfs/ might make some sense.
>>> Do you have a plan to integrate support for fs_locations to probe
>>> servers for alternate IP addresses? Would that be a userspace
>>> utility that would plug values into this new /sys API?
>
> Yeah, I would expect there to be a new utility to help with assigning
> new values. I haven't given any thought to using fs_locations yet, but
> it could probably work.
I could see a tool that performs an fs_locations from user space
and plugs that information into the kernel NFS client as needed.
Future work!
--
Chuck Lever
> On Jan 13, 2021, at 2:48 PM, Chuck Lever <[email protected]> wrote:
>
>> On Jan 13, 2021, at 2:23 PM, Anna Schumaker <[email protected]> wrote:
>>
>> On Tue, Jan 12, 2021 at 11:59 AM J. Bruce Fields <[email protected]> wrote:
>>>
>>> On Tue, Jan 12, 2021 at 08:09:09AM -0500, Chuck Lever wrote:
>>>> Hi Anna-
>>>>
>>>>> On Jan 11, 2021, at 4:41 PM, [email protected] wrote:
>>>>>
>>>>> From: Anna Schumaker <[email protected]>
>>>>>
>>>>> It's possible for an NFS server to go down but come back up with a
>>>>> different IP address. These patches provide a way for administrators to
>>>>> handle this issue by providing a new IP address for xprt sockets to
>>>>> connect to.
>>>>>
>>>>> This is a first draft of the code, so any thoughts or suggestions would
>>>>> be greatly appreciated!
>>>>
>>>> One implementation question, one future question.
>>>>
>>>> Would /sys/kernel/net be a little better? or /sys/kernel/sunrpc ?
>>
>> Possibly! I was trying to match /sys/fs/nfs, but I can definitely
>> change this if another location is better.
>
> Ah... since this is a supplement to the mount() interface, maybe
> placing this new API under /sys/fs/nfs/ might make some sense.
Or you could implement it with "-o remount,addr=new-address".
--
Chuck Lever
On Wed, Jan 13, 2021 at 9:18 PM Chuck Lever <[email protected]> wrote:
>
>
> > On Jan 13, 2021, at 2:48 PM, Chuck Lever <[email protected]> wrote:
> >
> >> On Jan 13, 2021, at 2:23 PM, Anna Schumaker <[email protected]> wrote:
> >>
> >> On Tue, Jan 12, 2021 at 11:59 AM J. Bruce Fields <[email protected]> wrote:
> >>>
> >>> On Tue, Jan 12, 2021 at 08:09:09AM -0500, Chuck Lever wrote:
> >>>> Hi Anna-
> >>>>
> >>>>> On Jan 11, 2021, at 4:41 PM, [email protected] wrote:
> >>>>>
> >>>>> From: Anna Schumaker <[email protected]>
> >>>>>
> >>>>> It's possible for an NFS server to go down but come back up with a
> >>>>> different IP address. These patches provide a way for administrators to
> >>>>> handle this issue by providing a new IP address for xprt sockets to
> >>>>> connect to.
> >>>>>
> >>>>> This is a first draft of the code, so any thoughts or suggestions would
> >>>>> be greatly appreciated!
> >>>>
> >>>> One implementation question, one future question.
> >>>>
> >>>> Would /sys/kernel/net be a little better? or /sys/kernel/sunrpc ?
> >>
> >> Possibly! I was trying to match /sys/fs/nfs, but I can definitely
> >> change this if another location is better.
> >
> > Ah... since this is a supplement to the mount() interface, maybe
> > placing this new API under /sys/fs/nfs/ might make some sense.
>
> Or you could implement it with "-o remount,addr=new-address".
A change of address is currently not allowed by the NFS because
multiple mounts might be sharing a superblock and change of one
mount's option would not be correct. The way things work from this new
mechanism is system wide and all mounts are affected.
>
>
> --
> Chuck Lever
>
>
>
> On Jan 14, 2021, at 3:29 PM, Olga Kornievskaia <[email protected]> wrote:
>
> On Wed, Jan 13, 2021 at 9:18 PM Chuck Lever <[email protected]> wrote:
>>
>>
>>> On Jan 13, 2021, at 2:48 PM, Chuck Lever <[email protected]> wrote:
>>>
>>>> On Jan 13, 2021, at 2:23 PM, Anna Schumaker <[email protected]> wrote:
>>>>
>>>> On Tue, Jan 12, 2021 at 11:59 AM J. Bruce Fields <[email protected]> wrote:
>>>>>
>>>>> On Tue, Jan 12, 2021 at 08:09:09AM -0500, Chuck Lever wrote:
>>>>>> Hi Anna-
>>>>>>
>>>>>>> On Jan 11, 2021, at 4:41 PM, [email protected] wrote:
>>>>>>>
>>>>>>> From: Anna Schumaker <[email protected]>
>>>>>>>
>>>>>>> It's possible for an NFS server to go down but come back up with a
>>>>>>> different IP address. These patches provide a way for administrators to
>>>>>>> handle this issue by providing a new IP address for xprt sockets to
>>>>>>> connect to.
>>>>>>>
>>>>>>> This is a first draft of the code, so any thoughts or suggestions would
>>>>>>> be greatly appreciated!
>>>>>>
>>>>>> One implementation question, one future question.
>>>>>>
>>>>>> Would /sys/kernel/net be a little better? or /sys/kernel/sunrpc ?
>>>>
>>>> Possibly! I was trying to match /sys/fs/nfs, but I can definitely
>>>> change this if another location is better.
>>>
>>> Ah... since this is a supplement to the mount() interface, maybe
>>> placing this new API under /sys/fs/nfs/ might make some sense.
>>
>> Or you could implement it with "-o remount,addr=new-address".
>
> A change of address is currently not allowed by the NFS because
> multiple mounts might be sharing a superblock and change of one
> mount's option would not be correct. The way things work from this new
> mechanism is system wide and all mounts are affected.
OK, well, if we're going with an API based on /sys that shows
underlying transport connections, is there a way to expose
whether the connection is established or closed? Maybe also
last traffic or last connect attempt?
Can it support RPC/RDMA connections too?
--
Chuck Lever