Skip to contentSkip to conversation

net/rds: handle zerocopy send cleanup before the message is queued

PATCH Applied Ren Wei 4 participants 5 messages May 1 – May 5 1 message not in the archive

Subsystems (4)
lore
Share & tools
lore

Ren Wei

author
#1 lore raw

PATCH 1/1 net/rds: handle zerocopy send cleanup before the message is queued

In reply to a message not in the archive

From: Nan Li <[email protected]>

A zerocopy send can fail after user pages have been pinned but before the message is attached to the sending socket.

The purge path currently infers zerocopy state from rm->m_rs, so an unqueued message can be cleaned up as if it owned normal payload pages. However, zerocopy ownership is really determined by the presence of op_mmp_znotifier, regardless of whether the message has reached the socket queue.

Capture op_mmp_znotifier up front in rds_message_purge() and use it as the cleanup discriminator. If the message is already associated with a socket, keep the existing completion path. Otherwise, drop the pinned page accounting directly and release the notifier before putting the payload pages.

This keeps early send failure cleanup consistent with the zerocopy lifetime rules without changing the normal queued completion path.

---
 net/rds/message.c | 20 +++++++++++++++-----
 1 file changed, 15 insertions(+), 5 deletions(-)
net/rds/message.c +15 -5
index eaa6f22601a4..25fedcb3cd00 100644
@@ -131,24 +131,34 @@ static void rds_rm_zerocopy_callback(struct rds_sock *rs,
  */
 static void rds_message_purge(struct rds_message *rm)
 {
+	struct rds_znotifier *znotifier;
 	unsigned long i, flags;
-	bool zcopy = false;
+	bool zcopy;
 
 	if (unlikely(test_bit(RDS_MSG_PAGEVEC, &rm->m_flags)))
 		return;
 
 	spin_lock_irqsave(&rm->m_rs_lock, flags);
+	znotifier = rm->data.op_mmp_znotifier;
+	rm->data.op_mmp_znotifier = NULL;
+	zcopy = !!znotifier;
+
 	if (rm->m_rs) {
 		struct rds_sock *rs = rm->m_rs;
 
-		if (rm->data.op_mmp_znotifier) {
-			zcopy = true;
-			rds_rm_zerocopy_callback(rs, rm->data.op_mmp_znotifier);
+		if (znotifier) {
+			rds_rm_zerocopy_callback(rs, znotifier);
 			rds_wake_sk_sleep(rs);
-			rm->data.op_mmp_znotifier = NULL;
 		}
 		sock_put(rds_rs_to_sk(rs));
 		rm->m_rs = NULL;
+	} else if (znotifier) {
+		/*
+		 * Zerocopy can fail before the message is queued on the
+		 * socket, so there is no rs to carry the notification.
+		 */
+		mm_unaccount_pinned_pages(&znotifier->z_mmp);
+		kfree(rds_info_from_znotifier(znotifier));
 	}
 	spin_unlock_irqrestore(&rm->m_rs_lock, flags);
 
Signature · 1 line
2.43.0
maintainer Reviewed-by on v1 1/1
#2 lore raw
Quoted 20 lines · Ren Wei

From: Nan Li <[email protected]>

A zerocopy send can fail after user pages have been pinned but before the message is attached to the sending socket.

The purge path currently infers zerocopy state from rm->m_rs, so an unqueued message can be cleaned up as if it owned normal payload pages. However, zerocopy ownership is really determined by the presence of op_mmp_znotifier, regardless of whether the message has reached the socket queue.

Capture op_mmp_znotifier up front in rds_message_purge() and use it as the cleanup discriminator. If the message is already associated with a socket, keep the existing completion path. Otherwise, drop the pinned page accounting directly and release the notifier before putting the payload pages.

This keeps early send failure cleanup consistent with the zerocopy lifetime rules without changing the normal queued completion path.

Quoted 10 lines

This fix looks fine to me. Thanks Ren Wei!

Allison

Quoted 48 lines
---
 net/rds/message.c | 20 +++++++++++++++-----
 1 file changed, 15 insertions(+), 5 deletions(-)
net/rds/message.c +15 -5
index eaa6f22601a4..25fedcb3cd00 100644
@@ -131,24 +131,34 @@ static void rds_rm_zerocopy_callback(struct rds_sock *rs,
  */
 static void rds_message_purge(struct rds_message *rm)
 {
+	struct rds_znotifier *znotifier;
 	unsigned long i, flags;
-	bool zcopy = false;
+	bool zcopy;
 
 	if (unlikely(test_bit(RDS_MSG_PAGEVEC, &rm->m_flags)))
 		return;
 
 	spin_lock_irqsave(&rm->m_rs_lock, flags);
+	znotifier = rm->data.op_mmp_znotifier;
+	rm->data.op_mmp_znotifier = NULL;
+	zcopy = !!znotifier;
+
 	if (rm->m_rs) {
 		struct rds_sock *rs = rm->m_rs;
 
-		if (rm->data.op_mmp_znotifier) {
-			zcopy = true;
-			rds_rm_zerocopy_callback(rs, rm->data.op_mmp_znotifier);
+		if (znotifier) {
+			rds_rm_zerocopy_callback(rs, znotifier);
 			rds_wake_sk_sleep(rs);
-			rm->data.op_mmp_znotifier = NULL;
 		}
 		sock_put(rds_rs_to_sk(rs));
 		rm->m_rs = NULL;
+	} else if (znotifier) {
+		/*
+		 * Zerocopy can fail before the message is queued on the
+		 * socket, so there is no rs to carry the notification.
+		 */
+		mm_unaccount_pinned_pages(&znotifier->z_mmp);
+		kfree(rds_info_from_znotifier(znotifier));
 	}
 	spin_unlock_irqrestore(&rm->m_rs_lock, flags);
 
maintainer
#3 lore raw
Quoted 32 lines · Allison Henderson
Quoted 20 lines · Ren Wei

From: Nan Li <[email protected]>

A zerocopy send can fail after user pages have been pinned but before the message is attached to the sending socket.

The purge path currently infers zerocopy state from rm->m_rs, so an unqueued message can be cleaned up as if it owned normal payload pages. However, zerocopy ownership is really determined by the presence of op_mmp_znotifier, regardless of whether the message has reached the socket queue.

Capture op_mmp_znotifier up front in rds_message_purge() and use it as the cleanup discriminator. If the message is already associated with a socket, keep the existing completion path. Otherwise, drop the pinned page accounting directly and release the notifier before putting the payload pages.

This keeps early send failure cleanup consistent with the zerocopy lifetime rules without changing the normal queued completion path.

Quoted 10 lines
Quoted 2 lines

This fix looks fine to me. Thanks Ren Wei!

Note that sashiko spotted more pre-existing problems in this area, please have a look:

https://sashiko.dev/#/patchset/d2ea98a6313d5467bac00f7c9fef8c7acddb9258.1777550074.git.tonanli66%40gmail.com

/P

#5 lore raw
In reply to Ren Wei
This patch was applied to netdev/net.git (main) by Paolo Abeni <[email protected]>:

Hello:

This patch was applied to netdev/net.git (main)
by Paolo Abeni <[email protected]>:

Quoted 5 lines · On Fri, 1 May 2026 09:08:44 +0800 you wrote:

From: Nan Li <[email protected]>

A zerocopy send can fail after user pages have been pinned but before the message is attached to the sending socket.

Quoted 7 lines

The purge path currently infers zerocopy state from rm->m_rs, so an unqueued message can be cleaned up as if it owned normal payload pages. However, zerocopy ownership is really determined by the presence of op_mmp_znotifier, regardless of whether the message has reached the socket queue.

[...]

Here is the summary with links:
- [net,1/1] net/rds: handle zerocopy send cleanup before the message is queued

    https://git.kernel.org/netdev/net/c/44b550d88b26

You are awesome, thank you!

Signature · 2 lines
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html
#4 lore raw
In reply to Paolo Abeni
Quoted 36 lines · Paolo Abeni
Quoted 32 lines · Allison Henderson
Quoted 20 lines · Ren Wei

From: Nan Li <[email protected]>

A zerocopy send can fail after user pages have been pinned but before the message is attached to the sending socket.

The purge path currently infers zerocopy state from rm->m_rs, so an unqueued message can be cleaned up as if it owned normal payload pages. However, zerocopy ownership is really determined by the presence of op_mmp_znotifier, regardless of whether the message has reached the socket queue.

Capture op_mmp_znotifier up front in rds_message_purge() and use it as the cleanup discriminator. If the message is already associated with a socket, keep the existing completion path. Otherwise, drop the pinned page accounting directly and release the notifier before putting the payload pages.

This keeps early send failure cleanup consistent with the zerocopy lifetime rules without changing the normal queued completion path.

Quoted 10 lines
Quoted 2 lines

This fix looks fine to me. Thanks Ren Wei!

Quoted 6 lines

Note that sashiko spotted more pre-existing problems in this area, please have a look:

https://sashiko.dev/#/patchset/d2ea98a6313d5467bac00f7c9fef8c7acddb9258.1777550074.git.tonanli66%40gmail.com

/P

Thanks for pointing this out. It looks like a valid catch, I will see if I can get a fix out today or tomorrow.

Thank you!
Allison

Quoted 1 line