On 11/17/20 1:48 AM, Shi Lei wrote:

...

Signed-off-by: Shi Lei <shi_lei@massclouds.com> ---    src/util/virnetlink.c | 25 +++++++++++++++++++++++++    src/util/virnetlink.h |  1 +    2 files changed, 26 insertions(+)

diff --git a/src/util/virnetlink.c b/src/util/virnetlink.c index fdd3a6a4..e191f63b 100644 --- a/src/util/virnetlink.c +++ b/src/util/virnetlink.c @@ -41,6 +41,18 @@ VIR_LOG_INIT("util.netlink");    #define NETLINK_ACK_TIMEOUT_S  (2*1000)       #if defined(WITH_LIBNL) + +/* + * VETH_INFO_PEER is defined in libnl, but it isn't exposed. + * We include it just like what iproute2 has done.

This is actually defined as a part of the basic netlink .h files, and is available in /usr/include/linux/veth.h. The file in iproute2 is just a copy of that file (in include/uapi/linux/veth.h). I'm not sure why they did that, possibly so they could build an ip binary containing the "new" (at the time) veth support on a machine that was still lacking support in the kernel header files?.

Anyway, I replaced this open-coded definition of VETH_INFO_* with

#include <linux/veth.h> and everything built with no problem. And I've checked and that file is present in the kernel headers at least as far back as kernel-3.18 (and we don't support anything older than that).

So, when you send V2 with the change I outlined in my review of Patch 2/2, could you replace this enum with the proper #include?

Thanks!

Okay. I also don't like introducing that enum. Thanks for letting me know  :-) Shi Lei

...

+ */ +enum { +    VETH_INFO_UNSPEC, +    VETH_INFO_PEER, + +    __VETH_INFO_MAX +}; +    /* State for a single netlink event handle */    struct virNetlinkEventHandle {        int watch; @@ -535,6 +547,19 @@ virNetlinkNewLink(const char *ifname,            NETLINK_MSG_NEST_END(nl_msg, infodata);        }    +    if (STREQ(type, "veth") && extra_args && extra_args->veth_peer) { +        struct nlattr *infoveth = NULL; + +        NETLINK_MSG_NEST_START(nl_msg, infodata, IFLA_INFO_DATA); +        NETLINK_MSG_NEST_START(nl_msg, infoveth, VETH_INFO_PEER); +        nlmsg_reserve(nl_msg, sizeof(struct ifinfomsg), 0); +        NETLINK_MSG_PUT(nl_msg, IFLA_IFNAME, +                        (strlen(extra_args->veth_peer) + 1), +                        extra_args->veth_peer); +        NETLINK_MSG_NEST_END(nl_msg, infoveth); +        NETLINK_MSG_NEST_END(nl_msg, infodata); +    } +        NETLINK_MSG_NEST_END(nl_msg, linkinfo);           if (extra_args) { diff --git a/src/util/virnetlink.h b/src/util/virnetlink.h index 7121eac4..7c4ed202 100644 --- a/src/util/virnetlink.h +++ b/src/util/virnetlink.h @@ -84,6 +84,7 @@ struct _virNetlinkNewLinkData {        const int *ifindex;             /* The index for the 'link' device */        const virMacAddr *mac;          /* The MAC address of the device */        const uint32_t *macvlan_mode;   /* The mode of macvlan */ +    const char *veth_peer;          /* The peer name for veth */    };       int virNetlinkNewLink(const char *ifname,

On 11/17/20 1:48 AM, Shi Lei wrote:

When netlink is supported, use netlink to create veth device pair rather than 'ip link' command.

Signed-off-by: Shi Lei <shi_lei@massclouds.com> --- src/util/virnetdevveth.c | 39 ++++++++++++++++++++++++++++++--------- 1 file changed, 30 insertions(+), 9 deletions(-)

diff --git a/src/util/virnetdevveth.c b/src/util/virnetdevveth.c index b3eee1af..996bf5dd 100644 --- a/src/util/virnetdevveth.c +++ b/src/util/virnetdevveth.c @@ -27,6 +27,7 @@ #include "virfile.h" #include "virstring.h" #include "virnetdev.h" +#include "virnetlink.h"

#define VIR_FROM_THIS VIR_FROM_NONE

@@ -116,7 +117,6 @@ int virNetDevVethCreate(char** veth1, char** veth2) for (i = 0; i < MAX_VETH_RETRIES; i++) { g_autofree char *veth1auto = NULL; g_autofree char *veth2auto = NULL; - g_autoptr(virCommand) cmd = NULL;

int status; if (!*veth1) { @@ -136,15 +136,32 @@ int virNetDevVethCreate(char** veth1, char** veth2) vethNum = veth2num + 1; }

- cmd = virCommandNew("ip"); - virCommandAddArgList(cmd, "link", "add", - *veth1 ? *veth1 : veth1auto, - "type", "veth", "peer", "name", - *veth2 ? *veth2 : veth2auto, - NULL); +#if defined(WITH_LIBNL) + { + int error = 0; + virNetlinkNewLinkData data = { + .veth_peer = *veth2 ? *veth2 : veth2auto, + };

- if (virCommandRun(cmd, &status) < 0) - goto cleanup; + status = virNetlinkNewLink(*veth1 ? *veth1 : veth1auto, + "veth", &data, &error); + if (status < 0) + goto cleanup; + } +#else + { + g_autoptr(virCommand) cmd = NULL; + cmd = virCommandNew("ip"); + virCommandAddArgList(cmd, "link", "add", + *veth1 ? *veth1 : veth1auto, + "type", "veth", "peer", "name", + *veth2 ? *veth2 : veth2auto, + NULL); + + if (virCommandRun(cmd, &status) < 0) + goto cleanup; + } +#endif /* WITH_LIBNL */

Although it isn't a hard/fast rule, we generally prefer to have complete functions within an #ifdefs rather then putting #ifdefs in the middle of a function. Could you put these bits into smaller functions, patterned like below, and then call the vir*Internal() functions from the existing functions? (other than that this looks fine, although I haven't actually tried *running* it yet :-)) #if defined(WITH_LIBNL) int virNetDevVethCreateInternal(char *veth1, char *veth2) {     int error;     virNetlinkNewLinkData data = { .veth_peer = veth2 };     return virNetlinkNewLink(veth1, "veth", &data, &error); } int virNetDevVethDeleteInternal(char *veth) {     return virNetlinkDelLink(veth, NULL); } #else int virNetDevVethCreateInternal(char *veth1, char *veth2) {     g_autoptr(virCommand) cmd = virCommandNew("ip");     virCommandAddArgList(cmd, "link", "add", veth1, veth2);     return virCommandRun(cmd, NULL); } int virNetDevVethDeleteInternal(char *veth) {     int status;     g_autoptr(virCommand) cmd = virCommandNewArgList("ip", "link", "del", veth, NULL);     if (virCommandRun(cmd, &status) < 0)         return -1;     if (status != 0) {         if (!virNetDevExists(veth)) {             VIR_DEBUG("Device %s already deleted (by kernel namespace cleanup)", veth);             return 0;         }         virReportError(VIR_ERR_INTERNAL_ERROR,                        _("Failed to delete veth device %s"), veth);         return -1;     }     return 0; }

if (status == 0) { if (veth1auto) { @@ -188,6 +205,9 @@ int virNetDevVethCreate(char** veth1, char** veth2) */ int virNetDevVethDelete(const char *veth) { +#if defined(WITH_LIBNL) + return virNetlinkDelLink(veth, NULL); +#else int status; g_autoptr(virCommand) cmd = virCommandNewArgList("ip", "link", "del", veth, NULL); @@ -206,4 +226,5 @@ int virNetDevVethDelete(const char *veth) }

return 0; +#endif /* WITH_LIBNL */ }

(Aside from this, it looks like veth interface naming could benefit from the same simplifications that I did to tap and macvtap awhile back (basically changing the auto-generated names to never re-use a name - commit 95089f481 and commit d7f38beb2e). But that isn't something for this series, just another thing to add to the to-do list :-))