Skip to contentSkip to conversation

vhost/vsock: reset only the h2g connections upon release

PATCH In review Jiyong Park 2 participants 5 messages Mar 10, 2022

Subsystems (7)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 vhost/vsock: reset only the h2g connections upon release

Filtering non-h2g connections out when determining orphaned connections. Otherwise, in a nested VM configuration, destroying the nested VM (which often involves the closing of /dev/vhost-vsock if there was h2g connections to the nested VM) kills not only the h2g connections, but also all existing g2h connections to the (outmost) host which are totally unrelated.

Tested: Executed the following steps on Cuttlefish (Android running on a VM) [1]: (1) Enter into an adb shell session - to have a g2h connection inside the VM, (2) open and then close /dev/vhost-vsock by exec 3< /dev/vhost-vsock && exec 3<&-, (3) observe that the adb session is not reset.

[1] https://android.googlesource.com/device/google/cuttlefish/

---
 drivers/vhost/vsock.c | 4 ++++
 1 file changed, 4 insertions(+)
drivers/vhost/vsock.c +4 -0
index 37f0b4274113..2f6d5d66f5ed 100644
@@ -722,6 +722,10 @@ static void vhost_vsock_reset_orphans(struct sock *sk)
 	 * executing.
 	 */
 
+	/* Only the h2g connections are reset */
+	if (vsk->transport != &vhost_transport.transport)
+		return;
+
 	/* If the peer is still valid, no need to reset connection */
 	if (vhost_vsock_get(vsk->remote_addr.svm_cid))
 		return;
Signature · 1 line
2.35.1.723.g4982287a31-goog
#1 lore raw

Hi Jiyong,

Quoted 38 lines · Jiyong Park

Filtering non-h2g connections out when determining orphaned connections. Otherwise, in a nested VM configuration, destroying the nested VM (which often involves the closing of /dev/vhost-vsock if there was h2g connections to the nested VM) kills not only the h2g connections, but also all existing g2h connections to the (outmost) host which are totally unrelated.

Tested: Executed the following steps on Cuttlefish (Android running on a VM) [1]: (1) Enter into an adb shell session - to have a g2h connection inside the VM, (2) open and then close /dev/vhost-vsock by exec 3< /dev/vhost-vsock && exec 3<&-, (3) observe that the adb session is not reset.

[1] https://android.googlesource.com/device/google/cuttlefish/

---
drivers/vhost/vsock.c | 4 ++++

1 file changed, 4 insertions(+)
drivers/vhost/vsock.c +4 -0
index 37f0b4274113..2f6d5d66f5ed 100644
@@ -722,6 +722,10 @@ static void vhost_vsock_reset_orphans(struct sock *sk)
 	 * executing.
 	 */
 
+	/* Only the h2g connections are reset */
+	if (vsk->transport != &vhost_transport.transport)
+		return;
+
 	/* If the peer is still valid, no need to reset connection */
 	if (vhost_vsock_get(vsk->remote_addr.svm_cid))
 		return;
Signature · 1 line
2.35.1.723.g4982287a31-goog

Thanks for your patch!

Yes, I see the problem and I think I introduced it with the multi-transports support (ooops).

So we should add this fixes tag:

Fixes c0cfa2d8a788 ("vsock: add multi-transports support")

IIUC the problem is for all transports that should only cycle on their own sockets. Indeed I think there is the same problem if the g2h driver will be unloaded (or a reset event is received after a VM migration), it will close all sockets of the nested h2g.

So I suggest a more generic solution, modifying vsock_for_each_connected_socket() like this (not tested):

net/vmw_vsock/af_vsock.c +7 -2
index 38baeb189d4e..f04abf662ec6 100644
@@ -334,7 +334,8 @@ void vsock_remove_sock(struct vsock_sock *vsk)
 }
 EXPORT_SYMBOL_GPL(vsock_remove_sock);
 
-void vsock_for_each_connected_socket(void (*fn)(struct sock *sk))
+void vsock_for_each_connected_socket(struct vsock_transport *transport,
+                                    void (*fn)(struct sock *sk))
 {
        int i;
 
@@ -343,8 +344,12 @@ void vsock_for_each_connected_socket(void (*fn)(struct sock *sk))
        for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++) {
                struct vsock_sock *vsk;
                list_for_each_entry(vsk, &vsock_connected_table[i],
-                                   connected_table)
+                                   connected_table) {
+                       if (vsk->transport != transport)
+                               continue;
+
                        fn(sk_vsock(vsk));
+               }
        }
 
 

And all transports that call it.

Thanks,
Stefano

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

#2 lore raw

Hi Stefano,

Quoted 91 lines · Stefano Garzarella

Hi Jiyong,

Quoted 36 lines · Jiyong Park

Filtering non-h2g connections out when determining orphaned connections. Otherwise, in a nested VM configuration, destroying the nested VM (which often involves the closing of /dev/vhost-vsock if there was h2g connections to the nested VM) kills not only the h2g connections, but also all existing g2h connections to the (outmost) host which are totally unrelated.

Tested: Executed the following steps on Cuttlefish (Android running on a VM) [1]: (1) Enter into an adb shell session - to have a g2h connection inside the VM, (2) open and then close /dev/vhost-vsock by exec 3< /dev/vhost-vsock && exec 3<&-, (3) observe that the adb session is not reset.

[1] https://android.googlesource.com/device/google/cuttlefish/

---
drivers/vhost/vsock.c | 4 ++++

1 file changed, 4 insertions(+)
drivers/vhost/vsock.c +4 -1
index 37f0b4274113..2f6d5d66f5ed 100644
@@ -722,6 +722,10 @@ static void vhost_vsock_reset_orphans(struct sock *sk)
       * executing.
       */
 
+      /* Only the h2g connections are reset */
+      if (vsk->transport != &vhost_transport.transport)
+              return;
+
      /* If the peer is still valid, no need to reset connection */
      if (vhost_vsock_get(vsk->remote_addr.svm_cid))
              return;
--
Quoted 2 lines

2.35.1.723.g4982287a31-goog

Thanks for your patch!

Yes, I see the problem and I think I introduced it with the multi-transports support (ooops).

So we should add this fixes tag:

Fixes c0cfa2d8a788 ("vsock: add multi-transports support")

IIUC the problem is for all transports that should only cycle on their own sockets. Indeed I think there is the same problem if the g2h driver will be unloaded (or a reset event is received after a VM migration), it will close all sockets of the nested h2g.

So I suggest a more generic solution, modifying vsock_for_each_connected_socket() like this (not tested):

net/vmw_vsock/af_vsock.c +7 -2
index 38baeb189d4e..f04abf662ec6 100644
@@ -334,7 +334,8 @@ void vsock_remove_sock(struct vsock_sock *vsk)
  }
  EXPORT_SYMBOL_GPL(vsock_remove_sock);
 
-void vsock_for_each_connected_socket(void (*fn)(struct sock *sk))
+void vsock_for_each_connected_socket(struct vsock_transport *transport,
+                                    void (*fn)(struct sock *sk))
  {
         int i;
 
@@ -343,8 +344,12 @@ void vsock_for_each_connected_socket(void (*fn)(struct sock *sk))
         for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++) {
                 struct vsock_sock *vsk;
                 list_for_each_entry(vsk, &vsock_connected_table[i],
-                                   connected_table)
+                                   connected_table) {
+                       if (vsk->transport != transport)
+                               continue;
+
                         fn(sk_vsock(vsk));
+               }
         }
 
 

And all transports that call it.

Quoted 3 lines

Thanks,
Stefano

Thanks for the suggestion, which looks much better. It actually worked well.

By the way, the suggested change will alter the kernel-module interface (KMI), which will make it difficult to land the change on older releases where we'd like to keep the KMI stable [1]. Would it be OK if we let the supplied function (fn) be responsible for checking the transport? I think that there, in the future,
might be a case where one needs to cycle over all sockets for inspection or so. I admit that this would be prone to error, though.

Please let me know what you think. I don't have a strong preference. I will submit a revision as you want.

[1] https://source.android.com/devices/architecture/kernel/generic-kernel-image#kmi-stability

#3 lore raw
Quoted 92 lines · Jiyong Park

Hi Stefano,

Quoted 89 lines · Stefano Garzarella

Hi Jiyong,

Quoted 36 lines · Jiyong Park

Filtering non-h2g connections out when determining orphaned connections. Otherwise, in a nested VM configuration, destroying the nested VM (which often involves the closing of /dev/vhost-vsock if there was h2g connections to the nested VM) kills not only the h2g connections, but also all existing g2h connections to the (outmost) host which are totally unrelated.

Tested: Executed the following steps on Cuttlefish (Android running on a VM) [1]: (1) Enter into an adb shell session - to have a g2h connection inside the VM, (2) open and then close /dev/vhost-vsock by exec 3< /dev/vhost-vsock && exec 3<&-, (3) observe that the adb session is not reset.

[1] https://android.googlesource.com/device/google/cuttlefish/

---
drivers/vhost/vsock.c | 4 ++++

1 file changed, 4 insertions(+)
drivers/vhost/vsock.c +4 -1
index 37f0b4274113..2f6d5d66f5ed 100644
@@ -722,6 +722,10 @@ static void vhost_vsock_reset_orphans(struct sock *sk)
       * executing.
       */
 
+      /* Only the h2g connections are reset */
+      if (vsk->transport != &vhost_transport.transport)
+              return;
+
      /* If the peer is still valid, no need to reset connection */
      if (vhost_vsock_get(vsk->remote_addr.svm_cid))
              return;
--
Quoted 2 lines

2.35.1.723.g4982287a31-goog

Thanks for your patch!

Yes, I see the problem and I think I introduced it with the multi-transports support (ooops).

So we should add this fixes tag:

Fixes c0cfa2d8a788 ("vsock: add multi-transports support")

IIUC the problem is for all transports that should only cycle on their own sockets. Indeed I think there is the same problem if the g2h driver will be unloaded (or a reset event is received after a VM migration), it will close all sockets of the nested h2g.

So I suggest a more generic solution, modifying vsock_for_each_connected_socket() like this (not tested):

net/vmw_vsock/af_vsock.c +7 -2
index 38baeb189d4e..f04abf662ec6 100644
@@ -334,7 +334,8 @@ void vsock_remove_sock(struct vsock_sock *vsk)
  }
  EXPORT_SYMBOL_GPL(vsock_remove_sock);
 
-void vsock_for_each_connected_socket(void (*fn)(struct sock *sk))
+void vsock_for_each_connected_socket(struct vsock_transport *transport,
+                                    void (*fn)(struct sock *sk))
  {
         int i;
 
@@ -343,8 +344,12 @@ void vsock_for_each_connected_socket(void (*fn)(struct sock *sk))
         for (i = 0; i < ARRAY_SIZE(vsock_connected_table); i++) {
                 struct vsock_sock *vsk;
                 list_for_each_entry(vsk, &vsock_connected_table[i],
-                                   connected_table)
+                                   connected_table) {
+                       if (vsk->transport != transport)
+                               continue;
+
                         fn(sk_vsock(vsk));
+               }
         }
 
 
Quoted 7 lines
Quoted 5 lines

And all transports that call it.

Thanks,
Stefano

Thanks for the suggestion, which looks much better. It actually worked well.

Thanks for trying this!

Quoted 11 lines

By the way, the suggested change will alter the kernel-module interface (KMI), which will make it difficult to land the change on older releases where we'd like to keep the KMI stable [1]. Would it be OK if we let the supplied function (fn) be responsible for checking the transport? I think that there, in the future,
might be a case where one needs to cycle over all sockets for inspection or so. I admit that this would be prone to error, though.

Please let me know what you think. I don't have a strong preference. I will submit a revision as you want.

I see your point, and it makes sense to keep KMI on stable branches, but mainline I would like to have the proposed approach since all transports use it to cycle on their sockets.

So we could do a series with 2 patches:
- Patch 1 fixes the problem in all transports by checking the transport in the callback (here we insert the fixes tag so we backport it to the stable branches)
- Patch 2 moves the check in vsock_for_each_connected_socket() removing it from callbacks so it is less prone to errors and we merge it only mainline

What do you think?

Thanks,
Stefano

#4 lore raw
Quoted 12 lines

I see your point, and it makes sense to keep KMI on stable branches, but mainline I would like to have the proposed approach since all transports use it to cycle on their sockets.

So we could do a series with 2 patches:
- Patch 1 fixes the problem in all transports by checking the transport in the callback (here we insert the fixes tag so we backport it to the stable branches)
- Patch 2 moves the check in vsock_for_each_connected_socket() removing it from callbacks so it is less prone to errors and we merge it only mainline

Quoted 5 lines

What do you think?

Thanks,
Stefano

That sounds like a plan. Will submit a new series soon.