| Message ID | 20170105160321.21786-2-berrange@redhat.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On 01/05/2017 10:03 AM, Daniel P. Berrange wrote: > Add a 'numeric' flag to the InetSocketAddress struct to allow the > caller to indicate that DNS should be skipped for the host/port > fields. This is useful if the caller knows the address is already > numeric and wants to guarantee no (potentially blocking) DNS > lookups are attempted. > > Signed-off-by: Daniel P. Berrange <berrange@redhat.com> > --- > qapi-schema.json | 5 +++++ > util/qemu-sockets.c | 3 +++ > 2 files changed, 8 insertions(+) > > diff --git a/qapi-schema.json b/qapi-schema.json > index a0d3b5d..d605c1e 100644 > --- a/qapi-schema.json > +++ b/qapi-schema.json > @@ -3070,6 +3070,10 @@ > # > # @port: port part of the address, or lowest port if @to is present > # > +# @numeric: #optional true if the host/port are guaranteed to be numeric, > +# false if name resolution should be attempted. Defaults to false. > +# Since 2.8 2.9, actually. I'm also not sure if Marc-Andre's work requires () brackets around the since designation on a per-member listing. > @@ -3084,6 +3088,7 @@ > 'data': { > 'host': 'str', > 'port': 'str', > + 'numeric': 'bool', In order to be optional, it must be spelled '*numeric'. > '*to': 'uint16', > '*ipv4': 'bool', > '*ipv6': 'bool' } } Thinking out loud: Do we even need a 'numeric' field? If we could create an alternate type that distinguishes between a 'str' (name, DNS resolution required) and an 'int' (numeric, skip DNS), then we don't need a field. Except that an 'int' for IPv4 addresses is awkward, and an 'int' for IPv6 addresses in insufficient - so even when the address is numeric, you STILL have to pass it in as 'str'. Okay, experiment failed, your interface seems like the correct thing to do. > diff --git a/util/qemu-sockets.c b/util/qemu-sockets.c > index fe1d07a..6f1c10a 100644 > --- a/util/qemu-sockets.c > +++ b/util/qemu-sockets.c > @@ -141,6 +141,9 @@ static int inet_listen_saddr(InetSocketAddress *saddr, > > memset(&ai,0, sizeof(ai)); > ai.ai_flags = AI_PASSIVE; > + if (saddr->numeric) { if (saddr->has_numeric && saddr->numeric) > + ai.ai_flags |= AI_NUMERICHOST | AI_NUMERICSERV; > + } > ai.ai_family = inet_ai_family_from_address(saddr, &err); > ai.ai_socktype = SOCK_STREAM; > >
On Thu, Jan 05, 2017 at 10:22:43AM -0600, Eric Blake wrote: > On 01/05/2017 10:03 AM, Daniel P. Berrange wrote: > > Add a 'numeric' flag to the InetSocketAddress struct to allow the > > caller to indicate that DNS should be skipped for the host/port > > fields. This is useful if the caller knows the address is already > > numeric and wants to guarantee no (potentially blocking) DNS > > lookups are attempted. > > > > Signed-off-by: Daniel P. Berrange <berrange@redhat.com> > > --- > > qapi-schema.json | 5 +++++ > > util/qemu-sockets.c | 3 +++ > > 2 files changed, 8 insertions(+) > > > > diff --git a/qapi-schema.json b/qapi-schema.json > > index a0d3b5d..d605c1e 100644 > > --- a/qapi-schema.json > > +++ b/qapi-schema.json > > @@ -3070,6 +3070,10 @@ > > # > > # @port: port part of the address, or lowest port if @to is present > > # > > +# @numeric: #optional true if the host/port are guaranteed to be numeric, > > +# false if name resolution should be attempted. Defaults to false. > > +# Since 2.8 > > 2.9, actually. I'm also not sure if Marc-Andre's work requires () > brackets around the since designation on a per-member listing. > > > @@ -3084,6 +3088,7 @@ > > 'data': { > > 'host': 'str', > > 'port': 'str', > > + 'numeric': 'bool', > > In order to be optional, it must be spelled '*numeric'. Sigh, yes. > > '*to': 'uint16', > > '*ipv4': 'bool', > > '*ipv6': 'bool' } } > > Thinking out loud: Do we even need a 'numeric' field? If we could > create an alternate type that distinguishes between a 'str' (name, DNS > resolution required) and an 'int' (numeric, skip DNS), then we don't > need a field. Except that an 'int' for IPv4 addresses is awkward, and > an 'int' for IPv6 addresses in insufficient - so even when the address > is numeric, you STILL have to pass it in as 'str'. Okay, experiment > failed, your interface seems like the correct thing to do. Basically think of this 'numeric' flag as being the same as the AI_NUMERIC flag to getaddrinfo. You're still using the string format as input so that you can reuse existing code for converting from user friendly string format to raw byte format. Regards, Daniel
diff --git a/qapi-schema.json b/qapi-schema.json index a0d3b5d..d605c1e 100644 --- a/qapi-schema.json +++ b/qapi-schema.json @@ -3070,6 +3070,10 @@ # # @port: port part of the address, or lowest port if @to is present # +# @numeric: #optional true if the host/port are guaranteed to be numeric, +# false if name resolution should be attempted. Defaults to false. +# Since 2.8 +# # @to: highest port to try # # @ipv4: whether to accept IPv4 addresses, default try both IPv4 and IPv6 @@ -3084,6 +3088,7 @@ 'data': { 'host': 'str', 'port': 'str', + 'numeric': 'bool', '*to': 'uint16', '*ipv4': 'bool', '*ipv6': 'bool' } } diff --git a/util/qemu-sockets.c b/util/qemu-sockets.c index fe1d07a..6f1c10a 100644 --- a/util/qemu-sockets.c +++ b/util/qemu-sockets.c @@ -141,6 +141,9 @@ static int inet_listen_saddr(InetSocketAddress *saddr, memset(&ai,0, sizeof(ai)); ai.ai_flags = AI_PASSIVE; + if (saddr->numeric) { + ai.ai_flags |= AI_NUMERICHOST | AI_NUMERICSERV; + } ai.ai_family = inet_ai_family_from_address(saddr, &err); ai.ai_socktype = SOCK_STREAM;
Add a 'numeric' flag to the InetSocketAddress struct to allow the caller to indicate that DNS should be skipped for the host/port fields. This is useful if the caller knows the address is already numeric and wants to guarantee no (potentially blocking) DNS lookups are attempted. Signed-off-by: Daniel P. Berrange <berrange@redhat.com> --- qapi-schema.json | 5 +++++ util/qemu-sockets.c | 3 +++ 2 files changed, 8 insertions(+)