[Qemu-devel] [PATCH] migration: setup bi-directional I/O channel for exec: protocol

[Qemu-devel] [PATCH] migration: setup bi-directional I/O channel for exec: protocol

[Daniel P. Berrange]

Historically the migration data channel has only needed to be
unidirectional. Thus the 'exec:' protocol was requesting an
I/O channel with O_RDONLY on incoming side, and O_WRONLY on
the outgoing side.

This is fine for classic migration, but if you then try to run
TLS over it, this fails because the TLS handshake requires a
bi-directional channel.

Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
---
 migration/exec.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/migration/exec.c b/migration/exec.c
index 9157721..aba9089 100644
--- a/migration/exec.c
+++ b/migration/exec.c
@@ -32,7 +32,7 @@ void exec_start_outgoing_migration(MigrationState *s, const char *command, Error
 
     trace_migration_exec_outgoing(command);
     ioc = QIO_CHANNEL(qio_channel_command_new_spawn(argv,
-                                                    O_WRONLY,
+                                                    O_RDWR,
                                                     errp));
     if (!ioc) {
         return;
@@ -59,7 +59,7 @@ void exec_start_incoming_migration(const char *command, Error **errp)
 
     trace_migration_exec_incoming(command);
     ioc = QIO_CHANNEL(qio_channel_command_new_spawn(argv,
-                                                    O_RDONLY,
+                                                    O_RDWR,
                                                     errp));
     if (!ioc) {
         return;
-- 
2.9.3

Re: [Qemu-devel] [PATCH] migration: setup bi-directional I/O channel for exec: protocol

[Juan Quintela]

"Daniel P. Berrange" <berrange@redhat.com> wrote:
> Historically the migration data channel has only needed to be
> unidirectional. Thus the 'exec:' protocol was requesting an
> I/O channel with O_RDONLY on incoming side, and O_WRONLY on
> the outgoing side.
>
> This is fine for classic migration, but if you then try to run
> TLS over it, this fails because the TLS handshake requires a
> bi-directional channel.
>
> Signed-off-by: Daniel P. Berrange <berrange@redhat.com>

Reviewed-by: Juan Quintela <quintela@redhat.com>

Actually, lots of people request that we have bi-directional
communication always that it is possible.