Re: [PATCH v4 2/8] migration: Converts uri parameter into 'MigrateAddress' struct

["Daniel P. Berrangé" <berrange@redhat.com>]

On Mon, May 15, 2023 at 01:55:59PM +0200, Juan Quintela wrote:
> Het Gala <het.gala@nutanix.com> wrote:
> v> Just so that, there is a wider attention, I will try to address and
> > discuss the comments from Daniel and Juan both here, as many of them
> > seems to be overlapping. I hope that is fine with the maintainers.
> >
> > On 15/05/23 3:42 pm, Daniel P. Berrangé wrote:
> >> On Fri, May 12, 2023 at 02:32:34PM +0000, Het Gala wrote:
> >>> This patch introduces code that can parse 'uri' string parameter and
> >>> spit out 'MigrateAddress' struct. All the required migration parameters
> >>> are stored in the struct.
> >>>
> >>> Suggested-by: Aravind Retnakaran <aravind.retnakaran@nutanix.com>
> >>> Signed-off-by: Het Gala <het.gala@nutanix.com>
> >>> ---
> >>>   migration/migration.c | 63 +++++++++++++++++++++++++++++++++++++++++--
> >>>   1 file changed, 61 insertions(+), 2 deletions(-)
> >>>
> >>> diff --git a/migration/migration.c b/migration/migration.c
> >>> index 0ee07802a5..a7e4e286aa 100644
> >>> --- a/migration/migration.c
> >>> +++ b/migration/migration.c
> >>> @@ -64,6 +64,7 @@
> >>>   #include "yank_functions.h"
> >>>   #include "sysemu/qtest.h"
> >>>   #include "options.h"
> >>> +#include "qemu/sockets.h"
> >>>     static NotifierList migration_state_notifiers =
> >>>       NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
> >>> @@ -408,13 +409,58 @@ void migrate_add_address(SocketAddress *address)
> >>>                         QAPI_CLONE(SocketAddress, address));
> >>>   }
> >>>   +static bool migrate_uri_parse(const char *uri,
> >>> +                              MigrateAddress **channel,
> >>> +                              Error **errp)
> >>> +{
> >>> +    Error *local_err = NULL;
> >>> +    MigrateAddress *addrs = g_new0(MigrateAddress, 1);
> >>> +    SocketAddress *saddr;
> >>> +    InetSocketAddress *isock = &addrs->u.rdma;
> >>> +    strList **tail = &addrs->u.exec.args;
> >>> +
> >>> +    if (strstart(uri, "exec:", NULL)) {
> >>> +        addrs->transport = MIGRATE_TRANSPORT_EXEC;
> >>> +        QAPI_LIST_APPEND(tail, g_strdup("/bin/sh"));
> >>> +        QAPI_LIST_APPEND(tail, g_strdup("-c"));
> >>> +        QAPI_LIST_APPEND(tail, g_strdup(uri + strlen("exec:")));
> >>> +    } else if (strstart(uri, "rdma:", NULL) &&
> >>> +               !inet_parse(isock, uri + strlen("rdma:"), errp)) {
> >>> +        addrs->transport = MIGRATE_TRANSPORT_RDMA;
> >> I would have this as
> >>
> >>      } else if (strstart(uri, "rdma:", NULL)) {
> >>          if (inet_parse(isock, uri + strlen("rdma:"), errp)) {
> >>              addrs->transport = MIGRATE_TRANSPORT_RDMA;
> >> 	}
> >>
> >> as IMHO it is bad practice to have control pass to the next
> >> else if clause when inet_parse() fails, as we know this is
> >> only an RDMA addr
> > Ack. I will change in the next patch.
> >> Also you need to use '&local_err' not 'errp' in the inet_parse
> >> call, otherwise the later code block for cleanup won't run.
> >
> > Yes, thanks for pointing it out Daniel. Will modify that.
> >
> > Also, Juan is of the opinion that we could omit 'local_error' variable
> > and try to address and free the memory there itself. For ex:
> >
> > if (saddr == NULL) {
> >     qapi_free_MigrateAddress(addrs);
> >     return false;
> > }
> >
> > Or, Daniel, can I also define here the variables like you suggested
> > down in the patch ? or is it used in some special case or I am missing
> > something ?
> >
> > g_autoptr(MigrateAddress) addrs = g_new0(MigrateAddress, 1);
> >
> > So we would not have to worry to free MigrateAddress struct.
> 
> https://blogs.gnome.org/desrt/2015/01/30/g_autoptr/
> 
> Yes, but that only happens for the cases where you want to always remove
> them.
> 
> >>> +    } else if (strstart(uri, "tcp:", NULL) ||
> >>> +                strstart(uri, "unix:", NULL) ||
> >>> +                strstart(uri, "vsock:", NULL) ||
> >>> +                strstart(uri, "fd:", NULL)) {
> >>> +        addrs->transport = MIGRATE_TRANSPORT_SOCKET;
> >>> +        saddr = socket_parse(uri, &local_err);
> >>> +        addrs->u.socket = *saddr;
> >> Protect with
> >>
> >>     if (saddr != NULL) {
> >>         addrs->u.socket = *saddr;
> >>     }
> >>
> >>> +    }
> >>> +
> >>> +    if (local_err) {
> >>> +        qapi_free_MigrateAddress(addrs);
> >>> +        qapi_free_SocketAddress(saddr);
> >>> +        qapi_free_InetSocketAddress(isock);
> >>> +        error_propagate(errp, local_err);
> >>> +        return false;
> >>> +    }
> >>> +
> >>> +    *channel = addrs;
> >>> +    return true;
> >>> +}
> >>> +
> >>>   static void qemu_start_incoming_migration(const char *uri, Error **errp)
> >>>   {
> >>>       const char *p = NULL;
> >>> +    MigrateAddress *channel = g_new0(MigrateAddress, 1);
> >> Avoid the later 'out:' cleanup block by using:
> >>
> >>    g_autoptr(MigrateAddress) channel = g_new0(MigrateAddress, 1);
> > Ack. I think this also solves the doubt raised by Juan "I wish, I
> > really wish, that there was a way to free things on error". Am I right
> > ?
> 
> No, that was the case where we have something like:
> 
> Thing *foo(void)
> {
>     OtherThing *bar = g_new0(OtherThing, 1)
> 
>     if (whatever) {
>         goto error;
>     }
>     if (whatever_else) {
>         goto error;
>     }
>     return bar;
> error:
>     g_free(bad);
>     return NULL;
> }
[> 
> See, we have to put the goto because we have to free it in all error
> paths.  Not in the non-error path.
> 
> If it is a pure local variable, i.e. never used after the function
> finishes, then g_autoptr is the right thing to do.

It is still better to use g_autoptr even in that case. You just need
to add in a call to g_steal_pointer in the success path. eg

 Thing *foo(void)
 {
     g_autoptr(OtherThing) bar = g_new0(OtherThing, 1)
 
     if (whatever) {
         return NULL;
     }
     if (whatever_else) {
         return NULL;
     }
     return g_steal_pointer(&bar);
 }


g_steal_pointer(&bar) is the equivalent of doing


    OtherThing *tmp = bar;
    bar = NULL;
    return tmp;

thus avoiding free'ing the pointer you're returning

With regards,
Daniel
-- 
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|