On Mon, 2025-02-17 at 17:57 -0500, Benjamin Marzinski wrote:
> On Fri, Feb 14, 2025 at 11:10:09PM +0100, Martin Wilck wrote:
> > When connecting to the multipathd socket, try the pathname socket
> > first, then the abstract socket. Fail only if both connection
> > attempts
> > fail.
> >
> > Signed-off-by: Martin Wilck <mwilck@xxxxxxxx>
> > ---
> > libmpathcmd/mpath_cmd.c | 13 ++++++++++---
> > 1 file changed, 10 insertions(+), 3 deletions(-)
> >
> > diff --git a/libmpathcmd/mpath_cmd.c b/libmpathcmd/mpath_cmd.c
> > index c7cf954..ba5bb31 100644
> > --- a/libmpathcmd/mpath_cmd.c
> > +++ b/libmpathcmd/mpath_cmd.c
> > @@ -102,7 +102,10 @@ int mpath_connect__(int nonblocking)
> > size_t len;
> > struct sockaddr_un addr;
> > int flags = 0;
> > + const char *names[2] = {PATHNAME_SOCKET, ABSTRACT_SOCKET};
> > + int name_idx = 0;
> >
> > +retry:
> > fd = socket(AF_LOCAL, SOCK_STREAM, 0);
> > if (fd == -1)
> > return -1;
> > @@ -113,13 +116,17 @@ int mpath_connect__(int nonblocking)
> > (void)fcntl(fd, F_SETFL,
> > flags|O_NONBLOCK);
> > }
> >
> > - len = mpath_fill_sockaddr__(&addr, ABSTRACT_SOCKET);
> > + len = mpath_fill_sockaddr__(&addr, names[name_idx]);
> > if (connect(fd, (struct sockaddr *)&addr, len) == -1) {
> > int err = errno;
> >
> > close(fd);
> > - errno = err;
> > - return -1;
> > + if (err == ECONNREFUSED && ++name_idx == 1)
>
> Most of the connect() return codes are things that could presumably
> be
> fixed by trying a different address (not the errors related to a
> problem
> with sockfd, but we just created the socket, so those seem pretty
> impossible). Is there a reason why we don't just retry on any error?
I've done some testing and when I was using a wrong address name, I
always got ECONNREFUSED. So I thought it might be cleaner to not
blindly assume retrying might help for other error codes. But I can see
your point, it's just a single retry and can't really hurt to just try
the other socket name.
I'll remove that condition.
Regards
Martin
