Skip to contentSkip to conversation

nfs4: allow server to change forechannel max_ops

PATCH In review J. Bruce Fields 3 participants 7 messages May 25, 2010 – Oct 2, 2010 1 message not in the archive

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

PATCH 1/1 nfs4: allow server to change forechannel max_ops

From: J. Bruce Fields <[email protected]>

Section 18.36.3 of rfc 5661 says that "For the fore channel, the server MAY change the requested value."

Also, there's no reason why the client would have to care if the server is willing to accept *more* operations per compound than the client requested.

---
 fs/nfs/nfs4proc.c |    1 -
 1 files changed, 0 insertions(+), 1 deletions(-)

On the other hand, if the server *decreases* max_ops on the forechannel, the client may need to do something. (Probably just fail for now.) Why aren't we checking for that case?

--b.

fs/nfs/nfs4proc.c +0 -1
index 071fced..a5a3690 100644
@@ -4880,7 +4880,6 @@ static int nfs4_verify_channel_attrs(struct nfs41_create_session_args *args,
 
 	ret |= _verify_fore_channel_attr(headerpadsz);
 	ret |= _verify_fore_channel_attr(max_resp_sz);
-	ret |= _verify_fore_channel_attr(max_ops);
 
 	ret |= _verify_back_channel_attr(headerpadsz);
 	ret |= _verify_back_channel_attr(max_rqst_sz);
Signature · 1 line
1.7.0.4
#1 lore raw
Quoted 25 lines · J. Bruce Fields

From: J. Bruce Fields <[email protected]>

Section 18.36.3 of rfc 5661 says that "For the fore channel, the server MAY change the requested value."

Also, there's no reason why the client would have to care if the server is willing to accept *more* operations per compound than the client requested.

---
 fs/nfs/nfs4proc.c |    1 -
 1 files changed, 0 insertions(+), 1 deletions(-)

On the other hand, if the server *decreases* max_ops on the forechannel, the client may need to do something. (Probably just fail for now.) Why aren't we checking for that case?

--b.

fs/nfs/nfs4proc.c +0 -0
index 071fced..a5a3690 100644

fs/nfs/nfs4proc.c static int nfs4_verify_channel_attrs(struct nfs41_create_session_args *args, · line 4885

Quoted 7 lines
fs/nfs/nfs4proc.c +0 -1
@@ -4880,7 +4880,6 @@ static int nfs4_verify_channel_attrs(struct nfs41_create_session_args *args,
 
 	ret |= _verify_fore_channel_attr(headerpadsz);
 	ret |= _verify_fore_channel_attr(max_resp_sz);
-	ret |= _verify_fore_channel_attr(max_ops);
 
 	ret |= _verify_back_channel_attr(headerpadsz);
 	ret |= _verify_back_channel_attr(max_rqst_sz);

Yes. That all looks wrong.

Can we please just get rid of that senseless macro, and open code those checks instead of the above patch? The current code is just pure obfuscation...

Trond

A message that is not in the archive (sent off-list or not yet received): 1274806624.11283.10.camel-rJ7iovZKK19ZJLDQqaL3InhyD016LWXt@public.gmane.org

#3 lore raw
Quoted 31 lines · Trond Myklebust
Quoted 30 lines · J. Bruce Fields

From: J. Bruce Fields <[email protected]>

Section 18.36.3 of rfc 5661 says that "For the fore channel, the server MAY change the requested value."

Also, there's no reason why the client would have to care if the server is willing to accept *more* operations per compound than the client requested.

---
 fs/nfs/nfs4proc.c |    1 -
 1 files changed, 0 insertions(+), 1 deletions(-)

On the other hand, if the server *decreases* max_ops on the forechannel, the client may need to do something. (Probably just fail for now.) Why aren't we checking for that case?

--b.

fs/nfs/nfs4proc.c +0 -1
index 071fced..a5a3690 100644
@@ -4880,7 +4880,6 @@ static int nfs4_verify_channel_attrs(struct nfs41_create_session_args *args,
 
 	ret |= _verify_fore_channel_attr(headerpadsz);
 	ret |= _verify_fore_channel_attr(max_resp_sz);
-	ret |= _verify_fore_channel_attr(max_ops);
 
Quoted 8 lines

fs/nfs/nfs4proc.c static int nfs4_verify_channel_attrs(struct nfs41_create_session_args *args, · line 4885

Quoted 2 lines
fs/nfs/nfs4proc.c +0 -0
 	ret |= _verify_back_channel_attr(headerpadsz);
 	ret |= _verify_back_channel_attr(max_rqst_sz);

Yes. That all looks wrong.

Can we please just get rid of that senseless macro, and open code those checks instead of the above patch? The current code is just pure obfuscation...

Sounds good to me. I'm hoping Andy can be roped into it....

--b.

#4 lore raw
Quoted 37 lines · J. Bruce Fields
Quoted 36 lines · Trond Myklebust
Quoted 25 lines · J. Bruce Fields

From: J. Bruce Fields <[email protected]>

Section 18.36.3 of rfc 5661 says that "For the fore channel, the server MAY change the requested value."

Also, there's no reason why the client would have to care if the server is willing to accept *more* operations per compound than the client requested.

---
 fs/nfs/nfs4proc.c |    1 -
 1 files changed, 0 insertions(+), 1 deletions(-)

On the other hand, if the server *decreases* max_ops on the forechannel, the client may need to do something. (Probably just fail for now.) Why aren't we checking for that case?

--b.

fs/nfs/nfs4proc.c +0 -0
index 071fced..a5a3690 100644

fs/nfs/nfs4proc.c static int nfs4_verify_channel_attrs(struct nfs41_create_session_args *args, · line 4885

Quoted 7 lines
fs/nfs/nfs4proc.c +0 -1
@@ -4880,7 +4880,6 @@ static int nfs4_verify_channel_attrs(struct nfs41_create_session_args *args,
 
 	ret |= _verify_fore_channel_attr(headerpadsz);
 	ret |= _verify_fore_channel_attr(max_resp_sz);
-	ret |= _verify_fore_channel_attr(max_ops);
 
 	ret |= _verify_back_channel_attr(headerpadsz);
 	ret |= _verify_back_channel_attr(max_rqst_sz);

Yes. That all looks wrong.

Quoted 5 lines
Quoted 3 lines

Can we please just get rid of that senseless macro, and open code those checks instead of the above patch? The current code is just pure obfuscation...

Sounds good to me. I'm hoping Andy can be roped into it....

Is anyone willing to work on this?

--b.

#5 lore raw

PATCH 1/1 nfs4: fix channel attribute sanity-checks

From: J. Bruce Fields <[email protected]>

The sanity checks here are incorrect; in the worst case they allow values that can later cause client crashes.

They're also over-reliant on the preprocessor.

---
 fs/nfs/nfs4proc.c |   75 +++++++++++++++++++++++++++++------------------------
 1 files changed, 41 insertions(+), 34 deletions(-)
Quoted 11 lines · J. Bruce Fields
Quoted 8 lines · J. Bruce Fields
Quoted 5 lines · Trond Myklebust

Yes. That all looks wrong.

Can we please just get rid of that senseless macro, and open code those checks instead of the above patch? The current code is just pure obfuscation...

Sounds good to me. I'm hoping Andy can be roped into it....

Is anyone willing to work on this?

OK, I cave. How about something like this?

Minimally tested. But that goes for the existing code, too.... A pynfs test or two might not be a bad idea.

--b.

fs/nfs/nfs4proc.c +41 -34
index 089da5b..eed13d6 100644
@@ -4914,49 +4914,56 @@ static void nfs4_init_channel_attrs(struct nfs41_create_session_args *args)
 		args->bc_attrs.max_reqs);
 }
 
-static int _verify_channel_attr(char *chan, char *attr_name, u32 sent, u32 rcvd)
+static int nfs4_verify_fore_channel_attrs(struct nfs41_create_session_args *args, struct nfs4_session *session)
 {
-	if (rcvd <= sent)
-		return 0;
-	printk(KERN_WARNING "%s: Session INVALID: %s channel %s increased. "
-		"sent=%u rcvd=%u\n", __func__, chan, attr_name, sent, rcvd);
-	return -EINVAL;
+	struct nfs4_channel_attrs *sent = &args->fc_attrs;
+	struct nfs4_channel_attrs *rcvd = &session->fc_attrs;
+
+	if (rcvd->headerpadsz > sent->headerpadsz)
+		return -EINVAL;
+	if (rcvd->max_resp_sz > sent->max_resp_sz)
+		return -EINVAL;
+	/*
+	 * Our requested max_ops is the minimum we need; we're not
+	 * prepared to break up compounds into smaller pieces than that.
+	 * So, not point even trying to continue if the server won't
+	 * cooperate:
+	 */
+	if (rcvd->max_ops < sent->max_ops)
+		return -EINVAL;
+	if (rcvd->max_reqs == 0)
+		return -EINVAL;
+	return 0;
 }
 
-#define _verify_fore_channel_attr(_name_) \
-	_verify_channel_attr("fore", #_name_, \
-			     args->fc_attrs._name_, \
-			     session->fc_attrs._name_)
+static int nfs4_verify_back_channel_attrs(struct nfs41_create_session_args *args, struct nfs4_session *session)
+{
+	struct nfs4_channel_attrs *sent = &args->bc_attrs;
+	struct nfs4_channel_attrs *rcvd = &session->bc_attrs;
 
-#define _verify_back_channel_attr(_name_) \
-	_verify_channel_attr("back", #_name_, \
-			     args->bc_attrs._name_, \
-			     session->bc_attrs._name_)
+	if (rcvd->max_rqst_sz > sent->max_rqst_sz)
+		return -EINVAL;
+	if (rcvd->max_resp_sz < sent->max_resp_sz)
+		return -EINVAL;
+	if (rcvd->max_resp_sz_cached > sent->max_resp_sz_cached)
+		return -EINVAL;
+	/* These would render the backchannel useless: */
+	if (rcvd->max_ops  == 0)
+		return -EINVAL;
+	if (rcvd->max_reqs == 0)
+		return -EINVAL;
+	return 0;
+}
 
-/*
- * The server is not allowed to increase the fore channel header pad size,
- * maximum response size, or maximum number of operations.
- *
- * The back channel attributes are only negotiatied down: We send what the
- * (back channel) server insists upon.
- */
 static int nfs4_verify_channel_attrs(struct nfs41_create_session_args *args,
 				     struct nfs4_session *session)
 {
-	int ret = 0;
-
-	ret |= _verify_fore_channel_attr(headerpadsz);
-	ret |= _verify_fore_channel_attr(max_resp_sz);
-	ret |= _verify_fore_channel_attr(max_ops);
-
-	ret |= _verify_back_channel_attr(headerpadsz);
-	ret |= _verify_back_channel_attr(max_rqst_sz);
-	ret |= _verify_back_channel_attr(max_resp_sz);
-	ret |= _verify_back_channel_attr(max_resp_sz_cached);
-	ret |= _verify_back_channel_attr(max_ops);
-	ret |= _verify_back_channel_attr(max_reqs);
+	int ret;
 
-	return ret;
+	ret = nfs4_verify_fore_channel_attrs(args, session);
+	if (ret)
+		return ret;
+	return nfs4_verify_back_channel_attrs(args, session);
 }
 
 static int _nfs4_proc_create_session(struct nfs_client *clp)
Signature · 1 line
1.7.0.4
#6 lore raw

Re: [PATCH] nfs4: fix channel attribute sanity-checks

J. Bruce Fields wrote:

+ * Our requested max_ops is the minimum we need; we're not
+ * prepared to break up compounds into smaller pieces than that.
+ * So, not point even trying to continue if the server won't
+ * cooperate:

"not" should be "no".

Also seems odd that max_ops is actually the minimum, but I guess the code is clear enough.

#7 lore raw

PATCH v2 1/1 nfs4: fix channel attribute sanity-checks

From: J. Bruce Fields <[email protected]>

The sanity checks here are incorrect; in the worst case they allow values that crash the client.

They're also over-reliant on the preprocessor.

---
 fs/nfs/nfs4proc.c |   75 +++++++++++++++++++++++++++++------------------------
 1 files changed, 41 insertions(+), 34 deletions(-)
Quoted 8 lines · Jim Rees

J. Bruce Fields wrote:

+ * Our requested max_ops is the minimum we need; we're not
+ * prepared to break up compounds into smaller pieces than that.
+ * So, not point even trying to continue if the server won't
+ * cooperate:

"not" should be "no".

Whoops, fixed.

Quoted 2 lines

Also seems odd that max_ops is actually the minimum, but I guess the code is clear enough.

Yeah. The naming's a little odd, at least; however this field was imagined being used when the spec was written, I don't think clients are likely to dynamically adjust their compounds somehow to optimize for the max_ops that the server gives them. The linux client behavior seems more likely: calculate the length of the longest compound the client needs to be able to send, send that as max_ops, error out if the server won't give at least much, and ignore the server if it grants more....

--b.

fs/nfs/nfs4proc.c +41 -34
index 089da5b..fdc79e5 100644
@@ -4914,49 +4914,56 @@ static void nfs4_init_channel_attrs(struct nfs41_create_session_args *args)
 		args->bc_attrs.max_reqs);
 }
 
-static int _verify_channel_attr(char *chan, char *attr_name, u32 sent, u32 rcvd)
+static int nfs4_verify_fore_channel_attrs(struct nfs41_create_session_args *args, struct nfs4_session *session)
 {
-	if (rcvd <= sent)
-		return 0;
-	printk(KERN_WARNING "%s: Session INVALID: %s channel %s increased. "
-		"sent=%u rcvd=%u\n", __func__, chan, attr_name, sent, rcvd);
-	return -EINVAL;
+	struct nfs4_channel_attrs *sent = &args->fc_attrs;
+	struct nfs4_channel_attrs *rcvd = &session->fc_attrs;
+
+	if (rcvd->headerpadsz > sent->headerpadsz)
+		return -EINVAL;
+	if (rcvd->max_resp_sz > sent->max_resp_sz)
+		return -EINVAL;
+	/*
+	 * Our requested max_ops is the minimum we need; we're not
+	 * prepared to break up compounds into smaller pieces than that.
+	 * So, no point even trying to continue if the server won't
+	 * cooperate:
+	 */
+	if (rcvd->max_ops < sent->max_ops)
+		return -EINVAL;
+	if (rcvd->max_reqs == 0)
+		return -EINVAL;
+	return 0;
 }
 
-#define _verify_fore_channel_attr(_name_) \
-	_verify_channel_attr("fore", #_name_, \
-			     args->fc_attrs._name_, \
-			     session->fc_attrs._name_)
+static int nfs4_verify_back_channel_attrs(struct nfs41_create_session_args *args, struct nfs4_session *session)
+{
+	struct nfs4_channel_attrs *sent = &args->bc_attrs;
+	struct nfs4_channel_attrs *rcvd = &session->bc_attrs;
 
-#define _verify_back_channel_attr(_name_) \
-	_verify_channel_attr("back", #_name_, \
-			     args->bc_attrs._name_, \
-			     session->bc_attrs._name_)
+	if (rcvd->max_rqst_sz > sent->max_rqst_sz)
+		return -EINVAL;
+	if (rcvd->max_resp_sz < sent->max_resp_sz)
+		return -EINVAL;
+	if (rcvd->max_resp_sz_cached > sent->max_resp_sz_cached)
+		return -EINVAL;
+	/* These would render the backchannel useless: */
+	if (rcvd->max_ops  == 0)
+		return -EINVAL;
+	if (rcvd->max_reqs == 0)
+		return -EINVAL;
+	return 0;
+}
 
-/*
- * The server is not allowed to increase the fore channel header pad size,
- * maximum response size, or maximum number of operations.
- *
- * The back channel attributes are only negotiatied down: We send what the
- * (back channel) server insists upon.
- */
 static int nfs4_verify_channel_attrs(struct nfs41_create_session_args *args,
 				     struct nfs4_session *session)
 {
-	int ret = 0;
-
-	ret |= _verify_fore_channel_attr(headerpadsz);
-	ret |= _verify_fore_channel_attr(max_resp_sz);
-	ret |= _verify_fore_channel_attr(max_ops);
-
-	ret |= _verify_back_channel_attr(headerpadsz);
-	ret |= _verify_back_channel_attr(max_rqst_sz);
-	ret |= _verify_back_channel_attr(max_resp_sz);
-	ret |= _verify_back_channel_attr(max_resp_sz_cached);
-	ret |= _verify_back_channel_attr(max_ops);
-	ret |= _verify_back_channel_attr(max_reqs);
+	int ret;
 
-	return ret;
+	ret = nfs4_verify_fore_channel_attrs(args, session);
+	if (ret)
+		return ret;
+	return nfs4_verify_back_channel_attrs(args, session);
 }
 
 static int _nfs4_proc_create_session(struct nfs_client *clp)
Signature · 1 line
1.7.0.4