On Fri, Feb 19, 2021 at 2:00 AM Daniel P. Berrangé <berra...@redhat.com>
wrote:
> On Thu, Feb 18, 2021 at 12:15:36PM -0800, Doug Evans wrote:
> > The parsing is moved into new function inet_parse_host_and_port.
> > This is done in preparation for using the function in net/slirp.c.
> >
> > Signed-off-by: Doug Evans <d...@google.com>
> > ---
> >
> > Changes from v3:
> > - this patch is new in v4
> > - provides new utility: inet_parse_host_and_port, updates inet_parse
> > to use it
> >
> > include/qemu/sockets.h | 3 ++
> > util/qemu-sockets.c | 94 +++++++++++++++++++++++++++++++-----------
> > 2 files changed, 72 insertions(+), 25 deletions(-)
> >
> > diff --git a/include/qemu/sockets.h b/include/qemu/sockets.h
> > index 7d1f813576..f720378a6b 100644
> > --- a/include/qemu/sockets.h
> > +++ b/include/qemu/sockets.h
> > @@ -31,6 +31,9 @@ int socket_set_fast_reuse(int fd);
> >
> > int inet_ai_family_from_address(InetSocketAddress *addr,
> > Error **errp);
> > +const char* inet_parse_host_and_port(const char* str, int terminator,
> > + char **addr, char **port, bool
> *is_v6,
> > + Error **errp);
> > int inet_parse(InetSocketAddress *addr, const char *str, Error **errp);
> > int inet_connect(const char *str, Error **errp);
> > int inet_connect_saddr(InetSocketAddress *saddr, Error **errp);
> > diff --git a/util/qemu-sockets.c b/util/qemu-sockets.c
> > index 8af0278f15..9fca7d9212 100644
> > --- a/util/qemu-sockets.c
> > +++ b/util/qemu-sockets.c
> > @@ -615,44 +615,88 @@ static int inet_parse_flag(const char *flagname,
> const char *optstr, bool *val,
> > return 0;
> > }
> >
> > -int inet_parse(InetSocketAddress *addr, const char *str, Error **errp)
> > +/*
> > + * Parse an inet host and port as "host:port<terminator>".
> > + * Terminator may be '\0'.
> > + * The syntax for ipv4 addresses is: address:port.
> > + * The syntax for ipv6 addresses is: [address]:port.
>
> It also supports
>
> "The syntax for hostnames is hostname:port
>
> > + * On success, returns a pointer to the terminator. Space for the
> address and
> > + * port is malloced and stored in *host, *port, the caller must free.
> > + * *is_v6 indicates whether the address is ipv4 or ipv6. If ipv6 then
> the
> > + * surrounding [] brackets are removed.
>
> When is_v6 is true, it indicates that a numeric ipv6 address was given.
> When false either a numberic ipv4 address or hostname was given.
>
> > + * On failure NULL is returned with the error stored in *errp.
> > + */
> > +const char* inet_parse_host_and_port(const char* str, int terminator,
> > + char **hostp, char **portp, bool
> *is_v6,
> > + Error **errp)
> > {
> > - const char *optstr, *h;
> > + const char *terminator_ptr = strchr(str, terminator);
> > + g_autofree char *buf = NULL;
> > char host[65];
> > char port[33];
> > - int to;
> > - int pos;
> > - char *begin;
> >
> > - memset(addr, 0, sizeof(*addr));
> > + if (terminator_ptr == NULL) {
> > + /* If the terminator isn't found then use the entire string. */
> > + terminator_ptr = str + strlen(str);
> > + }
> > + buf = g_strndup(str, terminator_ptr - str);
> >
> > - /* parse address */
> > - if (str[0] == ':') {
> > - /* no host given */
> > - host[0] = '\0';
> > - if (sscanf(str, ":%32[^,]%n", port, &pos) != 1) {
> > - error_setg(errp, "error parsing port in address '%s'", str);
> > - return -1;
> > - }
>
>
> > - } else if (str[0] == '[') {
> > + if (buf[0] == '[') {
> > /* IPv6 addr */
> > - if (sscanf(str, "[%64[^]]]:%32[^,]%n", host, port, &pos) != 2) {
> > - error_setg(errp, "error parsing IPv6 address '%s'", str);
> > - return -1;
> > + if (buf[1] == ']') {
> > + /* sscanf %[ doesn't recognize empty contents. */
> > + host[0] = '\0';
> > + if (sscanf(buf, "[]:%32s", port) != 1) {
> > + error_setg(errp, "error parsing IPv6 host:port '%s'",
> buf);
> > + return NULL;
> > + }
>
> This is introducing new functionality to the parser. Current callers
> let empty string ":port" be used for both ipv4 and ipv6, based
> on whether the flags ",ipv4[=on|off],ipv6[=on|off]" later follow.
>
We're creating a new utility subroutine: Let's decide what the API is for
it.
The fact that inet_parse is passed additional parameters to specify ipv4 vs
ipv6 is not something this new subroutine should care about.
I presume you want an explicit way to represent an empty ipv6 hostname
> to avoid changing semantics for existing slirp CLI args, where the
> existing ":port" exclusively means ipv4. IIC, this is also why you
> needed to introduce the "is_v6" flag, because any non-empty address
> can be reliably parsed without needing this flag.
>
Actually, no. The "is_v6" flag is needed because otherwise the caller has
no means (other than maybe subsequent grepping for "." vs ":") for knowing
whether str contained "address" or "[address]".
Plus, for my needs I don't need to support "[hostname]". If someone later
wants that supported that can be designed then.
Thus supporting "[hostname]" is not something I'm considering in this
patchset.
>
> This is reasonable, but any such functional change should be in a
> separate commit from refactoring.
>
> IOW, remove this and the is_v6 flag, and add them in a separate
> patch to explain to the need for new functionality in the parsing.
>
> Given that existing callers don't need to support "[]", we should
> not let that be parsed, unless the caller passing a "is_v6" pointer
> which is not NULL.
>
> > + } else {
> > + if (sscanf(buf, "[%64[^]]]:%32s", host, port) != 2) {
> > + error_setg(errp, "error parsing IPv6 host:port '%s'",
> buf);
> > + return NULL;
> > + }
> > }
> > } else {
> > - /* hostname or IPv4 addr */
> > - if (sscanf(str, "%64[^:]:%32[^,]%n", host, port, &pos) != 2) {
> > - error_setg(errp, "error parsing address '%s'", str);
> > - return -1;
> > + if (buf[0] == ':') {
> > + /* no host given */
> > + host[0] = '\0';
> > + if (sscanf(buf, ":%32s", port) != 1) {
> > + error_setg(errp, "error parsing host:port '%s'", buf);
> > + return NULL;
> > + }
>
> It would be preferreable if the parsing code was not re-ordered when
> extracting it. It doesn't look like a functional change, but I'm unsure
> why you moved it ?
>
> > + } else {
> > + /* hostname or IPv4 addr */
> > + if (sscanf(buf, "%64[^:]:%32s", host, port) != 2) {
> > + error_setg(errp, "error parsing host:port '%s'", buf);
> > + return NULL;
> > + }
> > }
> > }
> >
> > - addr->host = g_strdup(host);
> > - addr->port = g_strdup(port);
> > + *hostp = g_strdup(host);
> > + *portp = g_strdup(port);
> > + *is_v6 = buf[0] == '[';
> > +
> > + return terminator_ptr;
> > +}
> > +
> > +int inet_parse(InetSocketAddress *addr, const char *str, Error **errp)
> > +{
> > + const char *optstr, *h;
> > + bool is_v6;
> > + int to;
> > + int pos;
> > + char *begin;
> > +
> > + memset(addr, 0, sizeof(*addr));
> > +
> > + optstr = inet_parse_host_and_port(str, ',', &addr->host,
> &addr->port,
> > + &is_v6, errp);
>
> Just pass NULL since we don't need is_v6
>
> > + if (optstr == NULL) {
> > + return -1;
> > + }
> >
> > /* parse options */
> > - optstr = str + pos;
> > h = strstr(optstr, ",to=");
> > if (h) {
> > h += 4;
> > --
> > 2.30.0.617.g56c4b15f3c-goog
> >
I can certainly defer [] handling to a later patch series.
Splitting the patch into one with the is_v6 flag and one without is a lot
of work for little gain (zero IMO): When looking at
inet_parse_host_and_port() as its own utility subroutine, not providing the
caller with the means to distinguish whether str was "address:port" or
"[address]:port" is a poor API.
I can still revise patch to allow is_v6 being NULL though.
