Skip to contentSkip to conversation

Do not enter setup_callback_client when finding backchannel failed

PATCH Superseded Kinglong Mee 2 participants 8 messages Nov 27, 2013 – Dec 9, 2013

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Do not enter setup_callback_client when finding backchannel failed

If finding backchannel failed, nfsd should not enter setup_callback_client.

---
 fs/nfsd/nfs4callback.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
fs/nfsd/nfs4callback.c +2 -2
index 7f05cd1..755d6d6 100644
@@ -664,7 +664,7 @@ static int setup_callback_client(struct nfs4_client *clp, struct nfs4_cb_conn *c
 		args.authflavor = clp->cl_cred.cr_flavor;
 		clp->cl_cb_ident = conn->cb_ident;
 	} else {
-		if (!conn->cb_xprt)
+		if (!conn->cb_xprt || !ses)
 			return -EINVAL;
 		clp->cl_cb_conn.cb_xprt = conn->cb_xprt;
 		clp->cl_cb_session = ses;
@@ -982,7 +982,7 @@ static void nfsd4_process_cb_update(struct nfsd4_callback *cb)
 	}
 	spin_unlock(&clp->cl_lock);
 
-	err = setup_callback_client(clp, &conn, ses);
+	err = c ? setup_callback_client(clp, &conn, ses) : -ENOENT;
 	if (err) {
 		nfsd4_mark_cb_down(clp, err);
 		return;
Signature · 1 line
1.8.4.2
#1 lore raw
Quoted 12 lines · Kinglong Mee

If finding backchannel failed, nfsd should not enter setup_callback_client.

---
 fs/nfsd/nfs4callback.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
fs/nfsd/nfs4callback.c +0 -0
index 7f05cd1..755d6d6 100644

fs/nfsd/nfs4callback.c static int setup_callback_client(struct nfs4_client *clp, struct nfs4_cb_conn *c · line 667

Quoted 5 lines
fs/nfsd/nfs4callback.c +1 -1
@@ -664,7 +664,7 @@ static int setup_callback_client(struct nfs4_client *clp, struct nfs4_cb_conn *c
 		args.authflavor = clp->cl_cred.cr_flavor;
 		clp->cl_cb_ident = conn->cb_ident;
 	} else {
-		if (!conn->cb_xprt)
+		if (!conn->cb_xprt || !ses)

It looks to me like ses should be set whenenver conn is. Do you have reason to believe the contrary?

fs/nfsd/nfs4callback.c static void nfsd4_process_cb_update(struct nfsd4_callback *cb) · line 985

Quoted 9 lines
fs/nfsd/nfs4callback.c +1 -1
 			return -EINVAL;
 		clp->cl_cb_conn.cb_xprt = conn->cb_xprt;
 		clp->cl_cb_session = ses;
@@ -982,7 +982,7 @@ static void nfsd4_process_cb_update(struct nfsd4_callback *cb)
 	}
 	spin_unlock(&clp->cl_lock);
 
-	err = setup_callback_client(clp, &conn, ses);
+	err = c ? setup_callback_client(clp, &conn, ses) : -ENOENT;

But, whoops, yes, this looks like a good fix. (Have you hit this failure in practice?) But I think we should probably adopt the sort of control flow that's more idiomatic in the kernel:

	c = __nfsd4_find_backchannel(clp)
	if (!c)
		goto out_unlock;
	...
	spin_unlock(&clp->cl_lock);
	err = setup_callback_client(clp, &conn, ses);
	if (err)
		goto out_err;

etc.

--b.

Quoted 5 lines
fs/nfsd/nfs4callback.c +0 -0
 	if (err) {
 		nfsd4_mark_cb_down(clp, err);
 		return;
Signature · 1 line
1.8.4.2
#2 lore raw
Quoted 8 lines · J. Bruce Fields
Quoted 7 lines · Kinglong Mee

If finding backchannel failed, nfsd should not enter setup_callback_client.

---
 fs/nfsd/nfs4callback.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
Quoted 13 lines

fs/nfsd/nfs4callback.c static int setup_callback_client(struct nfs4_client *clp, struct nfs4_cb_conn *c · line 667

Quoted 10 lines
fs/nfsd/nfs4callback.c +1 -1
index 7f05cd1..755d6d6 100644
@@ -664,7 +664,7 @@ static int setup_callback_client(struct nfs4_client *clp, struct nfs4_cb_conn *c
 		args.authflavor = clp->cl_cred.cr_flavor;
 		clp->cl_cb_ident = conn->cb_ident;
 	} else {
-		if (!conn->cb_xprt)
+		if (!conn->cb_xprt || !ses)

It looks to me like ses should be set whenenver conn is. Do you have reason to believe the contrary?

No. It just a double check as conn->cb_xprt.
I will drop it in v2 of this path.

Quoted 8 lines
Quoted 8 lines
fs/nfsd/nfs4callback.c +0 -0
 			return -EINVAL;
 		clp->cl_cb_conn.cb_xprt = conn->cb_xprt;
 		clp->cl_cb_session = ses;
@@ -982,7 +982,7 @@ static void nfsd4_process_cb_update(struct nfsd4_callback *cb)
 	}
 	spin_unlock(&clp->cl_lock);
 
Quoted 5 lines

fs/nfsd/nfs4callback.c static void nfsd4_process_cb_update(struct nfsd4_callback *cb) · line 985

Quoted 2 lines
fs/nfsd/nfs4callback.c +1 -1
-	err = setup_callback_client(clp, &conn, ses);
+	err = c ? setup_callback_client(clp, &conn, ses) : -ENOENT;

But, whoops, yes, this looks like a good fix. (Have you hit this failure in practice?)

Yes. When destroying the last session from client, nfsd4_probe_callback_sync will be called in function nfsd4_destroy_session.

As that, callback client will be update in function nfsd4_process_cb_update, so __nfsd4_find_backchannel always failed with returning NULL. (Ps: Maybe needs another patch to avoid updating callback connect when destroying the last session ?)

For debugging, add message when finding backchannel failed.
Just testing mount/umount at client, the debug message will be print every umount.

fs/nfsd/nfs4callback.c +2 -0
index 7f05cd1..1be4c61 100644
@@ -979,6 +979,8 @@ static void nfsd4_process_cb_update(struct nfsd4_callback *cb)
                svc_xprt_get(c->cn_xprt);
                conn.cb_xprt = c->cn_xprt;
                ses = c->cn_session;
+       } else {
+               printk("%s: found connet failed.\n", __func__);
        }
        spin_unlock(&clp->cl_lock);
 
Quoted 12 lines

But I think we should probably adopt the sort of
control flow that's more idiomatic in the kernel:

	c = __nfsd4_find_backchannel(clp)
	if (!c)
		goto out_unlock;
	...
	spin_unlock(&clp->cl_lock);
	err = setup_callback_client(clp, &conn, ses);
	if (err)
		goto out_err;

OK, it is really better.

thanks,
Kinglong Mee

Quoted 10 lines

etc.

--b.

Quoted 5 lines
fs/nfsd/nfs4callback.c +0 -0
 	if (err) {
 		nfsd4_mark_cb_down(clp, err);
 		return;
Signature · 1 line
1.8.4.2
#7 lore raw

PATCH v2 1/1 Do not enter setup_callback_client when finding backchannel failed

In reply to J. Bruce Fields

Finding backchannel failed, nfsd should not enter setup_callback_client.

v2. drop ses checking in setup_callback_client and make code more clearer.

---
 fs/nfsd/nfs4callback.c | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)
fs/nfsd/nfs4callback.c +8 -4
index 7f05cd1..44d33ce 100644
@@ -975,11 +975,15 @@ static void nfsd4_process_cb_update(struct nfsd4_callback *cb)
 	clear_bit(NFSD4_CLIENT_CB_UPDATE, &clp->cl_flags);
 	memcpy(&conn, &cb->cb_clp->cl_cb_conn, sizeof(struct nfs4_cb_conn));
 	c = __nfsd4_find_backchannel(clp);
-	if (c) {
-		svc_xprt_get(c->cn_xprt);
-		conn.cb_xprt = c->cn_xprt;
-		ses = c->cn_session;
+	if (!c) {
+		spin_unlock(&clp->cl_lock);
+		nfsd4_mark_cb_down(clp, -ENOTCONN);
+		return;
 	}
+
+	svc_xprt_get(c->cn_xprt);
+	conn.cb_xprt = c->cn_xprt;
+	ses = c->cn_session;
 	spin_unlock(&clp->cl_lock);
 
 	err = setup_callback_client(clp, &conn, ses);
Signature · 1 line
1.8.4.2
#3 lore raw
In reply to Kinglong Mee
Quoted 40 lines · Kinglong Mee
Quoted 8 lines · J. Bruce Fields
Quoted 7 lines · Kinglong Mee

If finding backchannel failed, nfsd should not enter setup_callback_client.

---
 fs/nfsd/nfs4callback.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
Quoted 13 lines

fs/nfsd/nfs4callback.c static int setup_callback_client(struct nfs4_client *clp, struct nfs4_cb_conn *c · line 667

Quoted 10 lines
fs/nfsd/nfs4callback.c +1 -1
index 7f05cd1..755d6d6 100644
@@ -664,7 +664,7 @@ static int setup_callback_client(struct nfs4_client *clp, struct nfs4_cb_conn *c
 		args.authflavor = clp->cl_cred.cr_flavor;
 		clp->cl_cb_ident = conn->cb_ident;
 	} else {
-		if (!conn->cb_xprt)
+		if (!conn->cb_xprt || !ses)

It looks to me like ses should be set whenenver conn is. Do you have reason to believe the contrary?

No. It just a double check as conn->cb_xprt.
I will drop it in v2 of this path.

Quoted 8 lines
Quoted 8 lines
fs/nfsd/nfs4callback.c +0 -0
 			return -EINVAL;
 		clp->cl_cb_conn.cb_xprt = conn->cb_xprt;
 		clp->cl_cb_session = ses;
@@ -982,7 +982,7 @@ static void nfsd4_process_cb_update(struct nfsd4_callback *cb)
 	}
 	spin_unlock(&clp->cl_lock);
 
Quoted 5 lines

fs/nfsd/nfs4callback.c static void nfsd4_process_cb_update(struct nfsd4_callback *cb) · line 985

Quoted 2 lines
fs/nfsd/nfs4callback.c +1 -1
-	err = setup_callback_client(clp, &conn, ses);
+	err = c ? setup_callback_client(clp, &conn, ses) : -ENOENT;

But, whoops, yes, this looks like a good fix. (Have you hit this failure in practice?)

Quoted 5 lines

Yes. When destroying the last session from client, nfsd4_probe_callback_sync will be called in function nfsd4_destroy_session.

As that, callback client will be update in function nfsd4_process_cb_update, so __nfsd4_find_backchannel always failed with returning NULL.

On second thoughts, why is that actually a problem? setup_callback_client() will fail and you'll get an unnecessary printk from nfsd4_mark_cb_down, I guess. Is that the only problem?

--b.

Quoted 48 lines

(Ps: Maybe needs
another patch to avoid updating callback connect when destroying the last session ?)

For debugging, add message when finding backchannel failed.
Just testing mount/umount at client, the debug message will be print every umount.

fs/nfsd/nfs4callback.c +2 -0
index 7f05cd1..1be4c61 100644
@@ -979,6 +979,8 @@ static void nfsd4_process_cb_update(struct nfsd4_callback *cb)
                svc_xprt_get(c->cn_xprt);
                conn.cb_xprt = c->cn_xprt;
                ses = c->cn_session;
+       } else {
+               printk("%s: found connet failed.\n", __func__);
        }
        spin_unlock(&clp->cl_lock);
 
Quoted 12 lines

But I think we should probably adopt the sort of
control flow that's more idiomatic in the kernel:

	c = __nfsd4_find_backchannel(clp)
	if (!c)
		goto out_unlock;
	...
	spin_unlock(&clp->cl_lock);
	err = setup_callback_client(clp, &conn, ses);
	if (err)
		goto out_err;

OK, it is really better.

thanks,
Kinglong Mee

Quoted 10 lines

etc.

--b.

Quoted 5 lines
fs/nfsd/nfs4callback.c +0 -0
 	if (err) {
 		nfsd4_mark_cb_down(clp, err);
 		return;
Signature · 1 line
1.8.4.2
#4 lore raw
Quoted 44 lines · J. Bruce Fields
Quoted 43 lines · Kinglong Mee
Quoted 8 lines · J. Bruce Fields
Quoted 7 lines · Kinglong Mee

If finding backchannel failed, nfsd should not enter setup_callback_client.

---
 fs/nfsd/nfs4callback.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
Quoted 13 lines

fs/nfsd/nfs4callback.c static int setup_callback_client(struct nfs4_client *clp, struct nfs4_cb_conn *c · line 667

Quoted 10 lines
fs/nfsd/nfs4callback.c +1 -1
index 7f05cd1..755d6d6 100644
@@ -664,7 +664,7 @@ static int setup_callback_client(struct nfs4_client *clp, struct nfs4_cb_conn *c
 		args.authflavor = clp->cl_cred.cr_flavor;
 		clp->cl_cb_ident = conn->cb_ident;
 	} else {
-		if (!conn->cb_xprt)
+		if (!conn->cb_xprt || !ses)

It looks to me like ses should be set whenenver conn is. Do you have reason to believe the contrary?

No. It just a double check as conn->cb_xprt.
I will drop it in v2 of this path.

Quoted 8 lines
Quoted 8 lines
fs/nfsd/nfs4callback.c +0 -0
 			return -EINVAL;
 		clp->cl_cb_conn.cb_xprt = conn->cb_xprt;
 		clp->cl_cb_session = ses;
@@ -982,7 +982,7 @@ static void nfsd4_process_cb_update(struct nfsd4_callback *cb)
 	}
 	spin_unlock(&clp->cl_lock);
 
Quoted 5 lines

fs/nfsd/nfs4callback.c static void nfsd4_process_cb_update(struct nfsd4_callback *cb) · line 985

Quoted 2 lines
fs/nfsd/nfs4callback.c +1 -1
-	err = setup_callback_client(clp, &conn, ses);
+	err = c ? setup_callback_client(clp, &conn, ses) : -ENOENT;

But, whoops, yes, this looks like a good fix. (Have you hit this failure in practice?)

Yes. When destroying the last session from client, nfsd4_probe_callback_sync will be called in function nfsd4_destroy_session.

Quoted 6 lines
Quoted 2 lines

As that, callback client will be update in function nfsd4_process_cb_update, so __nfsd4_find_backchannel always failed with returning NULL.

On second thoughts, why is that actually a problem? setup_callback_client() will fail and you'll get an unnecessary printk from nfsd4_mark_cb_down, I guess.

Yes, that's right. setup_callback_client will failed with -EINVAL.
This patch just changes the logic, function returns immediately when finding backchannel failed, instead of enter setup_callback_client.

Quoted 1 line

Is that the only problem?

Yes, this path only focus the logic problem.

thanks,
Kinglong Mee

Quoted 49 lines
Quoted 48 lines

(Ps: Maybe needs
another patch to avoid updating callback connect when destroying the last session ?)

For debugging, add message when finding backchannel failed.
Just testing mount/umount at client, the debug message will be print every umount.

fs/nfsd/nfs4callback.c +2 -0
index 7f05cd1..1be4c61 100644
@@ -979,6 +979,8 @@ static void nfsd4_process_cb_update(struct nfsd4_callback *cb)
                svc_xprt_get(c->cn_xprt);
                conn.cb_xprt = c->cn_xprt;
                ses = c->cn_session;
+       } else {
+               printk("%s: found connet failed.\n", __func__);
        }
        spin_unlock(&clp->cl_lock);
 
Quoted 12 lines

But I think we should probably adopt the sort of
control flow that's more idiomatic in the kernel:

	c = __nfsd4_find_backchannel(clp)
	if (!c)
		goto out_unlock;
	...
	spin_unlock(&clp->cl_lock);
	err = setup_callback_client(clp, &conn, ses);
	if (err)
		goto out_err;

OK, it is really better.

thanks,
Kinglong Mee

Quoted 10 lines

etc.

--b.

Quoted 5 lines
fs/nfsd/nfs4callback.c +0 -0
 	if (err) {
 		nfsd4_mark_cb_down(clp, err);
 		return;
Signature · 1 line
1.8.4.2
#5 lore raw
Quoted 52 lines · Kinglong Mee
Quoted 44 lines · J. Bruce Fields
Quoted 43 lines · Kinglong Mee
Quoted 8 lines · J. Bruce Fields
Quoted 7 lines · Kinglong Mee

If finding backchannel failed, nfsd should not enter setup_callback_client.

---
 fs/nfsd/nfs4callback.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
Quoted 13 lines

fs/nfsd/nfs4callback.c static int setup_callback_client(struct nfs4_client *clp, struct nfs4_cb_conn *c · line 667

Quoted 10 lines
fs/nfsd/nfs4callback.c +1 -1
index 7f05cd1..755d6d6 100644
@@ -664,7 +664,7 @@ static int setup_callback_client(struct nfs4_client *clp, struct nfs4_cb_conn *c
 		args.authflavor = clp->cl_cred.cr_flavor;
 		clp->cl_cb_ident = conn->cb_ident;
 	} else {
-		if (!conn->cb_xprt)
+		if (!conn->cb_xprt || !ses)

It looks to me like ses should be set whenenver conn is. Do you have reason to believe the contrary?

No. It just a double check as conn->cb_xprt.
I will drop it in v2 of this path.

Quoted 8 lines
Quoted 8 lines
fs/nfsd/nfs4callback.c +0 -0
 			return -EINVAL;
 		clp->cl_cb_conn.cb_xprt = conn->cb_xprt;
 		clp->cl_cb_session = ses;
@@ -982,7 +982,7 @@ static void nfsd4_process_cb_update(struct nfsd4_callback *cb)
 	}
 	spin_unlock(&clp->cl_lock);
 
Quoted 5 lines

fs/nfsd/nfs4callback.c static void nfsd4_process_cb_update(struct nfsd4_callback *cb) · line 985

Quoted 2 lines
fs/nfsd/nfs4callback.c +1 -1
-	err = setup_callback_client(clp, &conn, ses);
+	err = c ? setup_callback_client(clp, &conn, ses) : -ENOENT;

But, whoops, yes, this looks like a good fix. (Have you hit this failure in practice?)

Yes. When destroying the last session from client, nfsd4_probe_callback_sync will be called in function nfsd4_destroy_session.

Quoted 6 lines
Quoted 2 lines

As that, callback client will be update in function nfsd4_process_cb_update, so __nfsd4_find_backchannel always failed with returning NULL.

On second thoughts, why is that actually a problem? setup_callback_client() will fail and you'll get an unnecessary printk from nfsd4_mark_cb_down, I guess.

Quoted 7 lines

Yes, that's right. setup_callback_client will failed with -EINVAL.
This patch just changes the logic, function returns immediately when finding backchannel failed, instead of enter setup_callback_client.

Quoted 1 line

Is that the only problem?

Yes, this path only focus the logic problem.

Apologies, I'm still not sure I understand:

As far as I can tell, the *only* way that your patch changes behavior is to change the error passed to nfsd4_mark_cb_path down.

Is that right, or am I missing some other problem which this fixes?

--b.

#6 lore raw

2013/12/7 J. Bruce Fields <[email protected]>:

Quoted 59 lines · Kinglong Mee
Quoted 58 lines · J. Bruce Fields
Quoted 44 lines · Kinglong Mee
Quoted 43 lines · J. Bruce Fields
Quoted 8 lines · Kinglong Mee
Quoted 7 lines

If finding backchannel failed, nfsd should not enter setup_callback_client.

---
 fs/nfsd/nfs4callback.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
Quoted 13 lines

fs/nfsd/nfs4callback.c static int setup_callback_client(struct nfs4_client *clp, struct nfs4_cb_conn *c · line 667

Quoted 10 lines
fs/nfsd/nfs4callback.c +1 -1
index 7f05cd1..755d6d6 100644
@@ -664,7 +664,7 @@ static int setup_callback_client(struct nfs4_client *clp, struct nfs4_cb_conn *c
                 args.authflavor = clp->cl_cred.cr_flavor;
                 clp->cl_cb_ident = conn->cb_ident;
         } else {
-                if (!conn->cb_xprt)
+                if (!conn->cb_xprt || !ses)

It looks to me like ses should be set whenenver conn is. Do you have reason to believe the contrary?

No. It just a double check as conn->cb_xprt.
I will drop it in v2 of this path.

Quoted 8 lines
Quoted 8 lines
fs/nfsd/nfs4callback.c +0 -0
                         return -EINVAL;
                 clp->cl_cb_conn.cb_xprt = conn->cb_xprt;
                 clp->cl_cb_session = ses;
@@ -982,7 +982,7 @@ static void nfsd4_process_cb_update(struct nfsd4_callback *cb)
         }
         spin_unlock(&clp->cl_lock);
 
Quoted 5 lines

fs/nfsd/nfs4callback.c static void nfsd4_process_cb_update(struct nfsd4_callback *cb) · line 985

Quoted 2 lines
fs/nfsd/nfs4callback.c +1 -1
-        err = setup_callback_client(clp, &conn, ses);
+        err = c ? setup_callback_client(clp, &conn, ses) : -ENOENT;

But, whoops, yes, this looks like a good fix. (Have you hit this failure in practice?)

Yes. When destroying the last session from client, nfsd4_probe_callback_sync will be called in function nfsd4_destroy_session.

Quoted 6 lines
Quoted 2 lines

As that, callback client will be update in function nfsd4_process_cb_update, so __nfsd4_find_backchannel always failed with returning NULL.

On second thoughts, why is that actually a problem? setup_callback_client() will fail and you'll get an unnecessary printk from nfsd4_mark_cb_down, I guess.

Yes, that's right. setup_callback_client will failed with -EINVAL.
This patch just changes the logic, function returns immediately when finding backchannel failed, instead of enter setup_callback_client.

Quoted 1 line

Is that the only problem?

Quoted 8 lines
Quoted 1 line

Yes, this path only focus the logic problem.

Apologies, I'm still not sure I understand:

As far as I can tell, the *only* way that your patch changes behavior is to change the error passed to nfsd4_mark_cb_path down.

Is that right, or am I missing some other problem which this fixes?

Sorry for the not clearly explan.
there are two behavior of this patch.
the first is as you said, change the error number.
the second is the logic, return immediately when finding session failed, not enter setup_callback_client.

thanks,
Kinglong Mee