macvlan: pass get_ts_info and SIOC[SG]HWTSTAMP ioctl to real device
Subsystems (2)
PATCH 1/1 macvlan: pass get_ts_info and SIOC[SG]HWTSTAMP ioctl to real device
Similiar to commit a6111d3c93d0 ("vlan: Pass SIOC[SG]HWTSTAMP ioctls to real device") and commit 37dd9255b2f6 ("vlan: Pass ethtool get_ts_info queries to real device."), add MACVlan HW ptp support.
--- drivers/net/macvlan.c | 48 +++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 48 insertions(+)
drivers/net/macvlan.c +48 -0
@@ -24,6 +24,7 @@#include <linux/notifier.h> #include <linux/netdevice.h> #include <linux/etherdevice.h> +#include <linux/net_tstamp.h> #include <linux/ethtool.h> #include <linux/if_arp.h> #include <linux/if_vlan.h>
@@ -34,6 +35,7 @@#include <net/rtnetlink.h> #include <net/xfrm.h> #include <linux/netpoll.h> +#include <linux/phy.h> #define MACVLAN_HASH_BITS 8#define MACVLAN_HASH_SIZE (1<<MACVLAN_HASH_BITS)
@@ -822,6 +824,30 @@ static int macvlan_change_mtu(struct net_device *dev, int new_mtu)return 0 ; } +static int macvlan_do_ioctl (struct net_device *dev,struct ifreq *ifr,int cmd) +{ +struct net_device *real_dev =macvlan_dev_real_dev (dev); +const struct net_device_ops *ops = real_dev->netdev_ops; +struct ifreq ifrr; +int err = -EOPNOTSUPP ; + +strncpy (ifrr.ifr_name, real_dev->name,IFNAMSIZ ); + ifrr.ifr_ifru = ifr->ifr_ifru; + +switch (cmd) { +case SIOCSHWTSTAMP : +case SIOCGHWTSTAMP : +if (netif_device_present (real_dev) && ops->ndo_do_ioctl) + err = ops->ndo_do_ioctl (real_dev, &ifrr, cmd); +break ; + } + +if (!err) + ifr->ifr_ifru = ifrr.ifr_ifru; + +return err; +} +/* * macvlan network devices have devices nesting below it and are a special * "super class" of normal network devices; split their locks off into a
@@ -1020,6 +1046,26 @@ static int macvlan_ethtool_get_link_ksettings(struct net_device *dev,return __ethtool_get_link_ksettings (vlan->lowerdev, cmd); } +static int macvlan_ethtool_get_ts_info (struct net_device *dev, +struct ethtool_ts_info *info) +{ +struct net_device *real_dev =macvlan_dev_real_dev (dev); +const struct ethtool_ops *ops = real_dev->ethtool_ops; +struct phy_device *phydev = real_dev->phydev; + +if (phydev && phydev->drv && phydev->drv->ts_info) { +return phydev->drv->ts_info (phydev, info); + }else if (ops->get_ts_info) { +return ops->get_ts_info (real_dev, info); + }else { + info->so_timestamping =SOF_TIMESTAMPING_RX_SOFTWARE | +SOF_TIMESTAMPING_SOFTWARE ; + info->phc_index =-1 ; + } + +return 0 ; +} +static netdev_features_t macvlan_fix_features (struct net_device *dev,netdev_features_t features) {
@@ -1094,6 +1140,7 @@ static const struct ethtool_ops macvlan_ethtool_ops = { .get_link = ethtool_op_get_link, .get_link_ksettings = macvlan_ethtool_get_link_ksettings, .get_drvinfo = macvlan_ethtool_get_drvinfo, + .get_ts_info = macvlan_ethtool_get_ts_info, };static const struct net_device_ops macvlan_netdev_ops = {
@@ -1103,6 +1150,7 @@ static const struct net_device_ops macvlan_netdev_ops = { .ndo_stop = macvlan_stop, .ndo_start_xmit = macvlan_start_xmit, .ndo_change_mtu = macvlan_change_mtu, + .ndo_do_ioctl = macvlan_do_ioctl, .ndo_fix_features = macvlan_fix_features, .ndo_change_rx_flags = macvlan_change_rx_flags, .ndo_set_mac_address = macvlan_set_mac_address,
Signature · 1 line
2.19.2
From: Hangbin Liu <[email protected]>
Date: Wed, 20 Mar 2019 10:23:33 +0800
Quoted 5 lines
Similiar to commit
a6111d3c93d0("vlan: Pass SIOC[SG]HWTSTAMP ioctls to real device") and commit37dd9255b2f6("vlan: Pass ethtoolget_ts_infoqueries to real device."), add MACVlan HW ptp support.Signed-off-by Hangbin Liu <[email protected]>
Applied.
Quoted 5 lines · Hangbin Liu
Similiar to commit
a6111d3c93d0("vlan: Pass SIOC[SG]HWTSTAMP ioctls to real device") and commit37dd9255b2f6("vlan: Pass ethtoolget_ts_infoqueries to real device."), add MACVlan HW ptp support.Signed-off-by Hangbin Liu <[email protected]>
Hi all,
For this patch. Jiri mentioned that running two ptp4l/phc2sys instances on two containers will not work. But I think the admin should avoid this scenario as we should not run two phc2sys instances on the same real device. In NET_ADMIN enabled containers, if a normal user which has mapped to root wants to run ptp4l, the admin need either add the device specifically (--device=/dev/ptp1) or give the container privileged permission (--privileged). So I think this should not be a security problem.
On the other hand, Miroslav pointed that with NET_ADMIN enabled in container, a normal user could be mapped to root and is able to change the real devices's rx filter via ioctl on macvlan, which may affect the other ptp process on host. ptp over vlan also has this issue, but macvlan is more frequently used in container. To prevent this issue, here are some options:
1. limit ioctl SIOCSHWTSTAMP to only enabling timestamping and switching to more general filters. The limitation could be only in container and leave init_net free.
2. reject SIOCSHWTSTAMP in container.
3. revert the patch. The vlan part is too late to revert.
So what do you think? which one do you prefer?
Thanks
Hangbin
Quoted 5 lines · Hangbin Liu
On the other hand, Miroslav pointed that with
NET_ADMINenabled in container, a normal user could be mapped to root and is able to change the real devices's rx filter via ioctl on macvlan, which may affect the other ptp process on host. ptp over vlan also has this issue, but macvlan is more frequently used in container.
If NET_ADMIN is enabled in the container, don't the host and container contend with each other for the physical interfaces anyhow?
(I'm not a container person, so forgive my ignorance.)
Thanks,
Richard
Quoted 2 lines · Richard Cochran
If
NET_ADMINis enabled in the container, don't the host and container contend with each other for the physical interfaces anyhow?
Physical interfaces are not a problem, as each interface can be only in a single net name space.
The problem here is this patch gives access to physical interface settings through a virtual interface layered on top of it. Whenever such thing is done, the virtual interface needs to provide a suitable way of moderating access to the shared resources, so the individual virtual interfaces do not affect each other. That's not what's being done here.
I think this patch is wrong.
Jiri
Quoted 6 lines · Jiri Benc
The problem here is this patch gives access to physical interface settings through a virtual interface layered on top of it. Whenever such thing is done, the virtual interface needs to provide a suitable way of moderating access to the shared resources, so the individual virtual interfaces do not affect each other. That's not what's being done here.
So I guess the macvlan should reject SIOCSHWTSTAMP but allow SIOCGHWTSTAMP.
Quoted 1 line
I think this patch is wrong.
But what about this statement:
ptp over vlan also has this issue
What is the issue with VLAN interfaces? Are these exportable to a container when the physical interface is not?
Thanks,
Richard
Quoted 10 lines · Richard Cochran
Quoted 6 lines · Jiri Benc
The problem here is this patch gives access to physical interface settings through a virtual interface layered on top of it. Whenever such thing is done, the virtual interface needs to provide a suitable way of moderating access to the shared resources, so the individual virtual interfaces do not affect each other. That's not what's being done here.
So I guess the macvlan should reject SIOCSHWTSTAMP but allow SIOCGHWTSTAMP.
Do you want to fix it only in container, like:
drivers/net/macvlan.c +2 -0
@@ -836,6 +836,8 @@ static int macvlan_do_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)switch (cmd) {case SIOCSHWTSTAMP : +if (!net_eq (dev_net (dev), &init_net)) +break ;case SIOCGHWTSTAMP :if (netif_device_present (real_dev) && ops->ndo_do_ioctl) err = ops->ndo_do_ioctl (real_dev, &ifrr, cmd);
Or just remove 'case SIOCSHWTSTAMP' directly?
And we should do the same on vlan, right?
Quoted 9 lines
Quoted 2 lines
I think this patch is wrong.
But what about this statement:
ptp over vlan also has this issue
What is the issue with VLAN interfaces? Are these exportable to a container when the physical interface is not?
Yes, if we attach a VLAN interface to container, a mapped user could also be able to modify the host's physical interface rx filter via ioctl.
Thanks
Hangbin
Quoted 10 lines · Richard Cochran
Quoted 6 lines · Jiri Benc
The problem here is this patch gives access to physical interface settings through a virtual interface layered on top of it. Whenever such thing is done, the virtual interface needs to provide a suitable way of moderating access to the shared resources, so the individual virtual interfaces do not affect each other. That's not what's being done here.
So I guess the macvlan should reject SIOCSHWTSTAMP but allow SIOCGHWTSTAMP.
FWIW, my suggestion was to limit what the SIOCSHWTSTAMP ioctl can do on the virtual interface. It could only enable HW timestamping or select a more general filter. A container could run a PTP clock if it had also access to the PHC device, or it could have the NET_ADMIN capability for other reasons, but it couldn't disable HW timestamping enabled by the host or other container.
If I understand it correctly, even without this ioctl a container can prevent the host or other containers from getting some of the HW timestamps by requesting TX timestamps at a high rate. I suspect the timestamping would need to be restricted to the real interface to fully protect it from applications having access to the virtual interfaces.
Signature · 1 line
Miroslav Lichvar
Hi Miroslav,
Quoted 5 lines · Miroslav Lichvar
Quoted 2 lines
So I guess the macvlan should reject SIOCSHWTSTAMP but allow SIOCGHWTSTAMP.
FWIW, my suggestion was to limit what the SIOCSHWTSTAMP ioctl can do on the virtual interface. It could only enable HW timestamping or
I think this is not enough as user could enable HWTSTAMP_FILTER_NONE.
Quoted 1 line
select a more general filter. A container could run a PTP clock if it
Do you have an idea about how to select a general filter? If we have enabled HWTSTAMP_FILTER_PTP_V2_L4_SYNC on host and a user in container want to enable HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ, then which one is more general?
Quoted 7 lines
had also access to the PHC device, or it could have the
NET_ADMINcapability for other reasons, but it couldn't disable HW timestamping enabled by the host or other container.If I understand it correctly, even without this ioctl a container can prevent the host or other containers from getting some of the HW timestamps by requesting TX timestamps at a high rate. I suspect the
Could traffic sharping/limitation fix it?
Quoted 3 lines
timestamping would need to be restricted to the real interface to fully protect it from applications having access to the virtual interfaces.
Thanks
Hangbin
Quoted 6 lines · Hangbin Liu
Quoted 1 line · Miroslav Lichvar
select a more general filter. A container could run a PTP clock if it
Do you have an idea about how to select a general filter? If we have enabled
HWTSTAMP_FILTER_PTP_V2_L4_SYNCon host and a user in container want to enableHWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ, then which one is more general?
In this case neither is a more general filter of the other. If V2_L4_SYNC is already selected, only the following filters could be selected on the macvlan interface:
HWTSTAMP_FILTER_PTP_V2_L4_SYNC, HWTSTAMP_FILTER_PTP_V2_L4_EVENT, HWTSTAMP_FILTER_PTP_V2_SYNC, HWTSTAMP_FILTER_PTP_V2_EVENT, HWTSTAMP_FILTER_ALL,
I think one way to check this would be to assign each filter a (16-bit?) value where the individual bits correspond to the message types and the newly selected filter would have to contain all bits of the old one.
Quoted 5 lines
Quoted 3 lines
If I understand it correctly, even without this ioctl a container can prevent the host or other containers from getting some of the HW timestamps by requesting TX timestamps at a high rate. I suspect the
Could traffic sharping/limitation fix it?
Yes, but it has to be specific to packets with TX timestamp requested. From what I have seen, TX timestamping may start to fail at just few tens of thousands of packets per second.
Signature · 1 line
Miroslav Lichvar
Quoted 9 lines · Miroslav Lichvar
Quoted 6 lines · Hangbin Liu
Quoted 1 line · Miroslav Lichvar
select a more general filter. A container could run a PTP clock if it
Do you have an idea about how to select a general filter? If we have enabled
HWTSTAMP_FILTER_PTP_V2_L4_SYNCon host and a user in container want to enableHWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ, then which one is more general?In this case neither is a more general filter of the other. If
Yes, that what I mean, some times it's hard to say which one is more general. like PTP_V2_L4_EVENT and PTP_V2_L2_SYNC.
Quoted 9 lines
V2_L4_SYNCis already selected, only the following filters could be selected on the macvlan interface:HWTSTAMP_FILTER_PTP_V2_L4_SYNC, HWTSTAMP_FILTER_PTP_V2_L4_EVENT, HWTSTAMP_FILTER_PTP_V2_SYNC, HWTSTAMP_FILTER_PTP_V2_EVENT, HWTSTAMP_FILTER_ALL,
Quoted 4 lines
I think one way to check this would be to assign each filter a (16-bit?) value where the individual bits correspond to the message types and the newly selected filter would have to contain all bits of the old one.
Just like I said, how to compare with different types.
Thanks
Hangbin
Quoted 10 lines · Hangbin Liu
Quoted 9 lines · Miroslav Lichvar
V2_L4_SYNCis already selected, only the following filters could be selected on the macvlan interface:HWTSTAMP_FILTER_PTP_V2_L4_SYNC, HWTSTAMP_FILTER_PTP_V2_L4_EVENT, HWTSTAMP_FILTER_PTP_V2_SYNC, HWTSTAMP_FILTER_PTP_V2_EVENT, HWTSTAMP_FILTER_ALL,
Quoted 6 lines
Quoted 4 lines
I think one way to check this would be to assign each filter a (16-bit?) value where the individual bits correspond to the message types and the newly selected filter would have to contain all bits of the old one.
Just like I said, how to compare with different types.
If those values I described above were in an array called ts_map indexed by the RX filter enum, I think the check could just be:
(ts_map[old_filter] & ts_map[new_filter]) == tsmap[old_filter]
The individual bits would correspond to:
PTP_V1_L4_SYNCPTP_V1_L4_DELAY_REQPTP_V2_L4_SYNCPTP_V2_L4_DELAY_REQPTP_V2_L2_SYNCPTP_V2_L2_DELAY_REQNTP_ALL
And the remaining RX filters would be combinations of those.
Signature · 1 line
Miroslav Lichvar
Quoted 15 lines · Miroslav Lichvar
If those values I described above were in an array called
ts_mapindexed by the RX filter enum, I think the check could just be:(ts_map[old_filter] & ts_map[new_filter]) == tsmap[old_filter]The individual bits would correspond to:
PTP_V1_L4_SYNCPTP_V1_L4_DELAY_REQPTP_V2_L4_SYNCPTP_V2_L4_DELAY_REQPTP_V2_L2_SYNCPTP_V2_L2_DELAY_REQNTP_ALL
Quoted 3 lines
And the remaining RX filters would be combinations of those.
Hi Miroslav, Richard,
Here is a draft patch with your idea. I haven't test it and it may has some issues. But the logic should looks like what you said. The copy_from/to_user is a little ugly, but I haven't come up with a more gentle way.
Would you please help have a look at it and see which way we should use? Drop SIOCSHWTSTAMP in container or add a filter on macvlan(maybe only in container)?
Thanks
Hangbin
drivers/net/macvlan.c +58 -1
@@ -824,18 +824,75 @@ static int macvlan_change_mtu(struct net_device *dev, int new_mtu)return 0 ; } +int check_rx_filter (unsigned int new_filter,unsigned int old_filter) +{ +u8 ts_map[HWTSTAMP_FILTER_NTP_ALL ]; + +memset (ts_map,0 ,sizeof (ts_map)); + + ts_map[HWTSTAMP_FILTER_PTP_V1_L4_SYNC -1 ] =0x01 ; + ts_map[HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ -1 ] =0x02 ; + ts_map[HWTSTAMP_FILTER_PTP_V1_L4_EVENT -1 ] =0x03 ; + + ts_map[HWTSTAMP_FILTER_PTP_V2_L4_SYNC -1 ] =0x11 ; + ts_map[HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ -1 ] =0x12 ; + ts_map[HWTSTAMP_FILTER_PTP_V2_L4_EVENT -1 ] =0x13 ; + + ts_map[HWTSTAMP_FILTER_PTP_V2_SYNC -1 ] =0x31 ; + ts_map[HWTSTAMP_FILTER_PTP_V2_DELAY_REQ -1 ] =0x32 ; + ts_map[HWTSTAMP_FILTER_PTP_V2_EVENT -1 ] =0x33 ; + + ts_map[HWTSTAMP_FILTER_NTP_ALL -1 ] =0xF0 ; + ts_map[HWTSTAMP_FILTER_ALL -1 ] =0xFF ; + +if ((ts_map[new_filter] & ts_map[old_filter]) == ts_map[old_filter]) +return 0 ; +else +return -EACCES ; +} +static int macvlan_do_ioctl (struct net_device *dev,struct ifreq *ifr,int cmd) {struct net_device *real_dev =macvlan_dev_real_dev (dev);const struct net_device_ops *ops = real_dev->netdev_ops; -struct ifreq ifrr; +unsigned int old_filter, new_filter, new_tx_type; +struct hwtstamp_config new_stmpconf, old_stmpconf;int err = -EOPNOTSUPP ; +struct ifreq ifrr; + +/* Get new rx_filter */ +if (copy_from_user (&new_stmpconf, ifr->ifr_data,sizeof (new_stmpconf))) { +return -EFAULT ; + }else { + new_tx_type = new_stmpconf.tx_type; + new_filter = new_stmpconf.rx_filter; + } +/* Get old rx_filter */ strncpy (ifrr.ifr_name, real_dev->name,IFNAMSIZ ); ifrr.ifr_ifru = ifr->ifr_ifru; +if (netif_device_present (real_dev) && ops->ndo_do_ioctl) + err = ops->ndo_do_ioctl (real_dev, &ifrr,SIOCGHWTSTAMP ); + +if (!err &©_from_user (&old_stmpconf, ifrr.ifr_data,sizeof (old_stmpconf))) + old_filter = old_stmpconf.rx_filter; +else +return err; + +/* Copy new data back */ +if (copy_to_user (ifrr.ifr_data, &new_stmpconf,sizeof (new_stmpconf))) +return -EFAULT ; +switch (cmd) {case SIOCSHWTSTAMP : +if (new_tx_type !=HWTSTAMP_TX_ON || + new_filter ==HWTSTAMP_FILTER_SOME ) +return err; + + err =check_rx_filter (new_filter, old_filter); +if (err) +break ;case SIOCGHWTSTAMP :if (netif_device_present (real_dev) && ops->ndo_do_ioctl) err = ops->ndo_do_ioctl (real_dev, &ifrr, cmd);
Signature · 1 line
2.19.2
Quoted 21 lines · Hangbin Liu
Quoted 15 lines · Miroslav Lichvar
If those values I described above were in an array called
ts_mapindexed by the RX filter enum, I think the check could just be:(ts_map[old_filter] & ts_map[new_filter]) == tsmap[old_filter]The individual bits would correspond to:
PTP_V1_L4_SYNCPTP_V1_L4_DELAY_REQPTP_V2_L4_SYNCPTP_V2_L4_DELAY_REQPTP_V2_L2_SYNCPTP_V2_L2_DELAY_REQNTP_ALLQuoted 3 lines
And the remaining RX filters would be combinations of those.
Hi Miroslav, Richard,
Quoted 7 lines
Here is a draft patch with your idea. I haven't test it and it may has some issues. But the logic should looks like what you said. The copy_from/to_user is a little ugly, but I haven't come up with a more gentle way.
Would you please help have a look at it and see which way we should use? Drop SIOCSHWTSTAMP in container or add a filter on macvlan(maybe only in container)?
Hi Richard,
Any suggestion?
Thanks
Hangbin
Quoted 85 lines
drivers/net/macvlan.c +58
-1@@ -824,18 +824,75 @@ static int macvlan_change_mtu(struct net_device *dev, int new_mtu)return 0 ; } +int check_rx_filter (unsigned int new_filter,unsigned int old_filter) +{ +u8 ts_map[HWTSTAMP_FILTER_NTP_ALL ]; + +memset (ts_map,0 ,sizeof (ts_map)); + + ts_map[HWTSTAMP_FILTER_PTP_V1_L4_SYNC -1 ] =0x01 ; + ts_map[HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ -1 ] =0x02 ; + ts_map[HWTSTAMP_FILTER_PTP_V1_L4_EVENT -1 ] =0x03 ; + + ts_map[HWTSTAMP_FILTER_PTP_V2_L4_SYNC -1 ] =0x11 ; + ts_map[HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ -1 ] =0x12 ; + ts_map[HWTSTAMP_FILTER_PTP_V2_L4_EVENT -1 ] =0x13 ; + + ts_map[HWTSTAMP_FILTER_PTP_V2_SYNC -1 ] =0x31 ; + ts_map[HWTSTAMP_FILTER_PTP_V2_DELAY_REQ -1 ] =0x32 ; + ts_map[HWTSTAMP_FILTER_PTP_V2_EVENT -1 ] =0x33 ; + + ts_map[HWTSTAMP_FILTER_NTP_ALL -1 ] =0xF0 ; + ts_map[HWTSTAMP_FILTER_ALL -1 ] =0xFF ; + +if ((ts_map[new_filter] & ts_map[old_filter]) == ts_map[old_filter]) +return 0 ; +else +return -EACCES ; +} +static int macvlan_do_ioctl (struct net_device *dev,struct ifreq *ifr,int cmd) {struct net_device *real_dev =macvlan_dev_real_dev (dev);const struct net_device_ops *ops = real_dev->netdev_ops; -struct ifreq ifrr; +unsigned int old_filter, new_filter, new_tx_type; +struct hwtstamp_config new_stmpconf, old_stmpconf;int err = -EOPNOTSUPP ; +struct ifreq ifrr; + +/* Get new rx_filter */ +if (copy_from_user (&new_stmpconf, ifr->ifr_data,sizeof (new_stmpconf))) { +return -EFAULT ; + }else { + new_tx_type = new_stmpconf.tx_type; + new_filter = new_stmpconf.rx_filter; + } +/* Get old rx_filter */ strncpy (ifrr.ifr_name, real_dev->name,IFNAMSIZ ); ifrr.ifr_ifru = ifr->ifr_ifru; +if (netif_device_present (real_dev) && ops->ndo_do_ioctl) + err = ops->ndo_do_ioctl (real_dev, &ifrr,SIOCGHWTSTAMP ); + +if (!err &©_from_user (&old_stmpconf, ifrr.ifr_data,sizeof (old_stmpconf))) + old_filter = old_stmpconf.rx_filter; +else +return err; + +/* Copy new data back */ +if (copy_to_user (ifrr.ifr_data, &new_stmpconf,sizeof (new_stmpconf))) +return -EFAULT ; +switch (cmd) {case SIOCSHWTSTAMP : +if (new_tx_type !=HWTSTAMP_TX_ON || + new_filter ==HWTSTAMP_FILTER_SOME ) +return err; + + err =check_rx_filter (new_filter, old_filter); +if (err) +break ;case SIOCGHWTSTAMP :if (netif_device_present (real_dev) && ops->ndo_do_ioctl) err = ops->ndo_do_ioctl (real_dev, &ifrr, cmd);Signature · 1 line
2.19.2
Quoted 3 lines · Hangbin Liu
Would you please help have a look at it and see which way we should use? Drop SIOCSHWTSTAMP in container or add a filter on macvlan(maybe only in container)?
I vote for dropping SIOCSHWTSTAMP altogether. Why? Because the filter idea means that the ioctl will magically succeed or fail, based on the unknowable state of the container's host. It is better IMHO to let the admin of the host set up HWTSTAMP globally (like with hwtstamp_ctl for example) and configure the apps appropriately (like with ptp4l --hwts_filter=check).
(BTW the patch has issues, but I'll let the advocates of the filter idea do the review ;)
Thanks,
Richard
Quoted 8 lines · Richard Cochran
Quoted 3 lines · Hangbin Liu
Would you please help have a look at it and see which way we should use? Drop SIOCSHWTSTAMP in container or add a filter on macvlan(maybe only in container)?
I vote for dropping SIOCSHWTSTAMP altogether. Why? Because the filter idea means that the ioctl will magically succeed or fail, based on the unknowable state of the container's host.
That's a good point. I agree that SIOCSHWTSTAMP always failing would be a less surprising behavior than failing only with some specific configurations.
Quoted 4 lines
It is better IMHO to
let the admin of the host set up HWTSTAMP globally (like withhwtstamp_ctlfor example) and configure the apps appropriately (like with ptp4l --hwts_filter=check).
Makes sense to me.
Some applications that support HW timestamping cannot do that currently (they call SIOCSHWTSTAMP even if nothing is changing), but it shouldn't be difficult to add support for this case.
Thanks,
Signature · 1 line
Miroslav Lichvar
Quoted 6 lines · Miroslav Lichvar
Quoted 5 lines · Richard Cochran
Quoted 3 lines · Hangbin Liu
Would you please help have a look at it and see which way we should use? Drop SIOCSHWTSTAMP in container or add a filter on macvlan(maybe only in container)?
Quoted 7 lines
Quoted 3 lines
I vote for dropping SIOCSHWTSTAMP altogether. Why? Because the filter idea means that the ioctl will magically succeed or fail, based on the unknowable state of the container's host.
That's a good point. I agree that SIOCSHWTSTAMP always failing would be a less surprising behavior than failing only with some specific configurations.
Thanks for the reply. As net-next is closed now. I will post the fix to net branch after merging finished.
Cheers
Hangbin
Quoted 11 lines · Hangbin Liu
Quoted 10 lines · Miroslav Lichvar
Quoted 8 lines · Richard Cochran
Quoted 3 lines · Hangbin Liu
Would you please help have a look at it and see which way we should use? Drop SIOCSHWTSTAMP in container or add a filter on macvlan(maybe only in container)?
I vote for dropping SIOCSHWTSTAMP altogether. Why? Because the filter idea means that the ioctl will magically succeed or fail, based on the unknowable state of the container's host.
Quoted 6 lines
Quoted 3 lines
That's a good point. I agree that SIOCSHWTSTAMP always failing would be a less surprising behavior than failing only with some specific configurations.
Thanks for the reply. As net-next is closed now. I will post the fix to net branch after merging finished.
net-next has been already merged into master and net so if it's a fix, you don't have to wait (and you shouldn't).
Michal Kubecek