12:18 p.m.
Trying to create a team device with name > 15 char results in a failure, but
a netdev with truncated name is created.
Netdevice "0123456789abcdef" not found.
teamd_init() failed.
Failed: No such device
114: 0123456789abcde@NONE: <BROADCAST,MULTICAST> mtu 1500 qdisc noop state DOWN mode
DEFAULT group default
link/ether 42:21:52:eb:37:a0 brd ff:ff:ff:ff:ff:ff
This patch fixes this issue by failing before a netdev with truncated name
is created.
Failed to create team device.
teamd_init() failed.
Failed: File name too long
Signed-off-by: Sridhar Samudrala <sridhar.samudrala(a)intel.com>
---
libteam/libteam.c | 3 +++
1 file changed, 3 insertions(+)
diff --git a/libteam/libteam.c b/libteam/libteam.c
index 4595081..ac187aa 100644
--- a/libteam/libteam.c
+++ b/libteam/libteam.c
@@ -458,6 +458,9 @@ static int do_create(struct team_handle *th, const char *team_name,
bool recreat
return -ENOMEM;
if (team_name) {
+ if (strlen(team_name) >= IFNAMSIZ)
+ return -ENAMETOOLONG;
+
rtnl_link_set_name(link, team_name);
if (recreate && team_ifname2ifindex(th, team_name)) {
--
2.1.0
12:28 p.m.
On 5/29/2015 10:18 AM, Sridhar Samudrala wrote:
Trying to create a team device with name > 15 char results in a
failure, but
a netdev with truncated name is created.
git commit removed the lines with #. The
missing lines are
# teamd -t 0123456789abcdef
Netdevice "0123456789abcdef" not found.
teamd_init() failed.
Failed: No such device
# ip link show
114: 0123456789abcde@NONE: <BROADCAST,MULTICAST> mtu 1500 qdisc
noop state DOWN mode DEFAULT group default
link/ether 42:21:52:eb:37:a0 brd ff:ff:ff:ff:ff:ff
This patch fixes this issue by failing before a netdev with truncated name
is created.
# ./teamd -t 0123456789abcdef
Failed to create team device.
teamd_init() failed.
Failed: File name too long
Signed-off-by: Sridhar Samudrala <sridhar.samudrala(a)intel.com>
---
libteam/libteam.c | 3 +++
1 file changed, 3 insertions(+)
diff --git a/libteam/libteam.c b/libteam/libteam.c
index 4595081..ac187aa 100644
--- a/libteam/libteam.c
+++ b/libteam/libteam.c
@@ -458,6 +458,9 @@ static int do_create(struct team_handle *th, const char *team_name,
bool recreat
return -ENOMEM;
if (team_name) {
+ if (strlen(team_name) >= IFNAMSIZ)
+ return -ENAMETOOLONG;
+
rtnl_link_set_name(link, team_name);
if (recreate && team_ifname2ifindex(th, team_name)) {
12:41 p.m.
On Fri, May 29, 2015 at 10:18:56AM -0700, Sridhar Samudrala wrote:
@@ -458,6 +458,9 @@ static int do_create(struct team_handle *th,
const char *team_name, bool recreat
return -ENOMEM;
if (team_name) {
+ if (strlen(team_name) >= IFNAMSIZ)
+ return -ENAMETOOLONG;
+
rtnl_link_set_name(link, team_name);
if (recreate && team_ifname2ifindex(th, team_name)) {
It looks odd to me that the libnl exposes an API such that it receives
its opaque object and a string and Team is responsible for checking if
the string fits into libnl purposes.
I'd assume that Team passes any string as interface name and libnl
would either do some checking or try and then report errors back to
the caller.
fbl
4:03 p.m.
On 5/29/2015 10:41 AM, Flavio Leitner wrote:
On Fri, May 29, 2015 at 10:18:56AM -0700, Sridhar Samudrala wrote:
> @@ -458,6 +458,9 @@ static int do_create(struct team_handle *th, const char
*team_name, bool recreat
> return -ENOMEM;
>
> if (team_name) {
> + if (strlen(team_name) >= IFNAMSIZ)
> + return -ENAMETOOLONG;
> +
> rtnl_link_set_name(link, team_name);
>
> if (recreate && team_ifname2ifindex(th, team_name)) {
It looks odd to me that the libnl exposes an API such that it receives
its opaque object and a string and Team is responsible for checking if
the string fits into libnl purposes.
I see some discussion to add this check to
rtnl_link_set_name(), but it
looks like that
patch didn't make it to libnl.
http://lists.infradead.org/pipermail/libnl/2014-January/001223.html
http://lists.infradead.org/pipermail/libnl/2014-January/001224.html
I'd assume that Team passes any string as interface name and libnl
would either do some checking or try and then report errors back to
the caller.
fbl
4:12 a.m.
Fri, May 29, 2015 at 11:03:54PM CEST, sridhar.samudrala(a)intel.com wrote:
On 5/29/2015 10:41 AM, Flavio Leitner wrote:
>On Fri, May 29, 2015 at 10:18:56AM -0700, Sridhar Samudrala wrote:
>>@@ -458,6 +458,9 @@ static int do_create(struct team_handle *th, const char
*team_name, bool recreat
>> return -ENOMEM;
>> if (team_name) {
>>+ if (strlen(team_name) >= IFNAMSIZ)
>>+ return -ENAMETOOLONG;
>>+
>> rtnl_link_set_name(link, team_name);
>> if (recreate && team_ifname2ifindex(th, team_name)) {
>It looks odd to me that the libnl exposes an API such that it receives
>its opaque object and a string and Team is responsible for checking if
>the string fits into libnl purposes.
I see some discussion to add this check to rtnl_link_set_name(), but it looks
like that
patch didn't make it to libnl.
http://lists.infradead.org/pipermail/libnl/2014-January/001223.html
http://lists.infradead.org/pipermail/libnl/2014-January/001224.html
The problem with libnl fix is that it breaks abi. So in the meantime,
I'm applying this to fix in libteam. Thanks!
3262
days inactive
3263
days old
libteam@lists.fedorahosted.org
4 comments
3 participants
participants (3)
-
Flavio Leitner -
Jiri Pirko -
Samudrala, Sridhar