For troubleshooting, it is useful to show the callback address and state,
even though we do have this equivalent info inside Chuck's ftrace patches.
Note there is a show_cb_state() inside fs/nfsd/trace.h and this code
has a similar function. It may be better to consolidate these two
if these additions are ok for nfsd client info, but not sure where
a good header is to place it - do we need a new file, maybe
fs/nfsd/nfs4callback.h?
Dave Wysochanski (1):
nfsd4: Expose the callback address and state of each NFS4 client
fs/nfsd/nfs4state.c | 17 +++++++++++++++++
1 file changed, 17 insertions(+)
--
1.8.3.1
In addition to the client's address, display the callback channel
state and address in the 'info' file.
Signed-off-by: Dave Wysochanski <[email protected]>
---
fs/nfsd/nfs4state.c | 17 +++++++++++++++++
1 file changed, 17 insertions(+)
diff --git a/fs/nfsd/nfs4state.c b/fs/nfsd/nfs4state.c
index 49c052243b5c..89a7cada334d 100644
--- a/fs/nfsd/nfs4state.c
+++ b/fs/nfsd/nfs4state.c
@@ -2357,6 +2357,21 @@ static void seq_quote_mem(struct seq_file *m, char *data, int len)
seq_printf(m, "\"");
}
+static const char *cb_state_str(int state)
+{
+ switch (state) {
+ case NFSD4_CB_UP:
+ return "UP";
+ case NFSD4_CB_UNKNOWN:
+ return "UNKNOWN";
+ case NFSD4_CB_DOWN:
+ return "DOWN";
+ case NFSD4_CB_FAULT:
+ return "FAULT";
+ }
+ return "UNDEFINED";
+}
+
static int client_info_show(struct seq_file *m, void *v)
{
struct inode *inode = m->private;
@@ -2385,6 +2400,8 @@ static int client_info_show(struct seq_file *m, void *v)
seq_printf(m, "\nImplementation time: [%lld, %ld]\n",
clp->cl_nii_time.tv_sec, clp->cl_nii_time.tv_nsec);
}
+ seq_printf(m, "callback state: %s\n", cb_state_str(clp->cl_cb_state));
+ seq_printf(m, "callback address: %pISpc\n", &clp->cl_cb_conn.cb_addr);
drop_client(clp);
return 0;
--
1.8.3.1
On Fri, May 14, 2021 at 09:30:40AM -0400, Dave Wysochanski wrote:
> For troubleshooting, it is useful to show the callback address and state,
> even though we do have this equivalent info inside Chuck's ftrace patches.
Good idea.
> Note there is a show_cb_state() inside fs/nfsd/trace.h and this code
> has a similar function. It may be better to consolidate these two
> if these additions are ok for nfsd client info, but not sure where
> a good header is to place it - do we need a new file, maybe
> fs/nfsd/nfs4callback.h?
nfs4state.c already includes trace.h, do we need anything more?
I'll admit I've just been adding things wherever seems expedient for a
while, so there may be some more logical way to organize nfsd headers.
--b.
>
> Dave Wysochanski (1):
> nfsd4: Expose the callback address and state of each NFS4 client
>
> fs/nfsd/nfs4state.c | 17 +++++++++++++++++
> 1 file changed, 17 insertions(+)
>
> --
> 1.8.3.1
Howdy Dave-
> On May 14, 2021, at 9:30 AM, Dave Wysochanski <[email protected]> wrote:
>
> In addition to the client's address, display the callback channel
> state and address in the 'info' file.
>
> Signed-off-by: Dave Wysochanski <[email protected]>
> ---
> fs/nfsd/nfs4state.c | 17 +++++++++++++++++
> 1 file changed, 17 insertions(+)
>
> diff --git a/fs/nfsd/nfs4state.c b/fs/nfsd/nfs4state.c
> index 49c052243b5c..89a7cada334d 100644
> --- a/fs/nfsd/nfs4state.c
> +++ b/fs/nfsd/nfs4state.c
> @@ -2357,6 +2357,21 @@ static void seq_quote_mem(struct seq_file *m, char *data, int len)
> seq_printf(m, "\"");
> }
>
> +static const char *cb_state_str(int state)
> +{
> + switch (state) {
> + case NFSD4_CB_UP:
> + return "UP";
> + case NFSD4_CB_UNKNOWN:
> + return "UNKNOWN";
> + case NFSD4_CB_DOWN:
> + return "DOWN";
> + case NFSD4_CB_FAULT:
> + return "FAULT";
No objection to the addition of this information. Style nit:
the "case" and "switch" lines should have the same amount of
indentation.
> + }
> + return "UNDEFINED";
> +}
> +
> static int client_info_show(struct seq_file *m, void *v)
> {
> struct inode *inode = m->private;
> @@ -2385,6 +2400,8 @@ static int client_info_show(struct seq_file *m, void *v)
> seq_printf(m, "\nImplementation time: [%lld, %ld]\n",
> clp->cl_nii_time.tv_sec, clp->cl_nii_time.tv_nsec);
> }
> + seq_printf(m, "callback state: %s\n", cb_state_str(clp->cl_cb_state));
> + seq_printf(m, "callback address: %pISpc\n", &clp->cl_cb_conn.cb_addr);
> drop_client(clp);
>
> return 0;
> --
> 1.8.3.1
>
--
Chuck Lever
On Fri, May 14, 2021 at 11:06 AM Chuck Lever III <[email protected]> wrote:
>
> Howdy Dave-
>
> > On May 14, 2021, at 9:30 AM, Dave Wysochanski <[email protected]> wrote:
> >
> > In addition to the client's address, display the callback channel
> > state and address in the 'info' file.
> >
> > Signed-off-by: Dave Wysochanski <[email protected]>
> > ---
> > fs/nfsd/nfs4state.c | 17 +++++++++++++++++
> > 1 file changed, 17 insertions(+)
> >
> > diff --git a/fs/nfsd/nfs4state.c b/fs/nfsd/nfs4state.c
> > index 49c052243b5c..89a7cada334d 100644
> > --- a/fs/nfsd/nfs4state.c
> > +++ b/fs/nfsd/nfs4state.c
> > @@ -2357,6 +2357,21 @@ static void seq_quote_mem(struct seq_file *m, char *data, int len)
> > seq_printf(m, "\"");
> > }
> >
> > +static const char *cb_state_str(int state)
> > +{
> > + switch (state) {
> > + case NFSD4_CB_UP:
> > + return "UP";
> > + case NFSD4_CB_UNKNOWN:
> > + return "UNKNOWN";
> > + case NFSD4_CB_DOWN:
> > + return "DOWN";
> > + case NFSD4_CB_FAULT:
> > + return "FAULT";
>
> No objection to the addition of this information. Style nit:
> the "case" and "switch" lines should have the same amount of
> indentation.
>
whoops! Thanks Chuck - I'll be sure to run checkpatch and fix it up in v2.
>
> > + }
> > + return "UNDEFINED";
> > +}
> > +
> > static int client_info_show(struct seq_file *m, void *v)
> > {
> > struct inode *inode = m->private;
> > @@ -2385,6 +2400,8 @@ static int client_info_show(struct seq_file *m, void *v)
> > seq_printf(m, "\nImplementation time: [%lld, %ld]\n",
> > clp->cl_nii_time.tv_sec, clp->cl_nii_time.tv_nsec);
> > }
> > + seq_printf(m, "callback state: %s\n", cb_state_str(clp->cl_cb_state));
> > + seq_printf(m, "callback address: %pISpc\n", &clp->cl_cb_conn.cb_addr);
> > drop_client(clp);
> >
> > return 0;
> > --
> > 1.8.3.1
> >
>
> --
> Chuck Lever
>
>
>
On Fri, May 14, 2021 at 10:15 AM Bruce Fields <[email protected]> wrote:
>
> On Fri, May 14, 2021 at 09:30:40AM -0400, Dave Wysochanski wrote:
> > For troubleshooting, it is useful to show the callback address and state,
> > even though we do have this equivalent info inside Chuck's ftrace patches.
>
> Good idea.
>
> > Note there is a show_cb_state() inside fs/nfsd/trace.h and this code
> > has a similar function. It may be better to consolidate these two
> > if these additions are ok for nfsd client info, but not sure where
> > a good header is to place it - do we need a new file, maybe
> > fs/nfsd/nfs4callback.h?
>
> nfs4state.c already includes trace.h, do we need anything more?
>
Probably not. I am testing a renamed function (I find that
"<typename>2str" is more common in the kernel) "cb_state2str"
defined in fs/nfsd/trace.c and declaration in fs/nfsd/trace.h
If that makes sense I'll send a v2.
> I'll admit I've just been adding things wherever seems expedient for a
> while, so there may be some more logical way to organize nfsd headers.
>
> --b.
>
> >
> > Dave Wysochanski (1):
> > nfsd4: Expose the callback address and state of each NFS4 client
> >
> > fs/nfsd/nfs4state.c | 17 +++++++++++++++++
> > 1 file changed, 17 insertions(+)
> >
> > --
> > 1.8.3.1
>