[patch net-next v5 0/9] devlink: introduce notifications filtering

[patch net-next v5 0/9] devlink: introduce notifications filtering

[Jiri Pirko]

From: Jiri Pirko <jiri@nvidia.com>

Currently the user listening on a socket for devlink notifications
gets always all messages for all existing devlink instances and objects,
even if he is interested only in one of those. That may cause
unnecessary overhead on setups with thousands of instances present.

User is currently able to narrow down the devlink objects replies
to dump commands by specifying select attributes.

Allow similar approach for notifications providing user a new
notify-filter-set command to select attributes with values
the notification message has to match. In that case, it is delivered
to the socket.

Note that the filtering is done per-socket, so multiple users may
specify different selection of attributes with values.

This patchset initially introduces support for following attributes:
DEVLINK_ATTR_BUS_NAME
DEVLINK_ATTR_DEV_NAME
DEVLINK_ATTR_PORT_INDEX

Patches #1 - #4 are preparations in devlink code, patch #3 is
                an optimization done on the way.
Patches #5 - #7 are preparations in netlink and generic netlink code.
Patch #8 is the main one in this set implementing of
         the notify-filter-set command and the actual
         per-socket filtering.
Patch #9 extends the infrastructure allowing to filter according
         to a port index.

Example:
$ devlink mon port pci/0000:08:00.0/32768
[port,new] pci/0000:08:00.0/32768: type notset flavour pcisf controller 0 pfnum 0 sfnum 107 splittable false
  function:
    hw_addr 00:00:00:00:00:00 state inactive opstate detached roce enable
[port,new] pci/0000:08:00.0/32768: type eth flavour pcisf controller 0 pfnum 0 sfnum 107 splittable false
  function:
    hw_addr 00:00:00:00:00:00 state inactive opstate detached roce enable
[port,new] pci/0000:08:00.0/32768: type eth netdev eth3 flavour pcisf controller 0 pfnum 0 sfnum 107 splittable false
  function:
    hw_addr 00:00:00:00:00:00 state inactive opstate detached roce enable
[port,new] pci/0000:08:00.0/32768: type eth netdev eth3 flavour pcisf controller 0 pfnum 0 sfnum 107 splittable false
  function:
    hw_addr 00:00:00:00:00:00 state inactive opstate detached roce enable
[port,new] pci/0000:08:00.0/32768: type eth flavour pcisf controller 0 pfnum 0 sfnum 107 splittable false
  function:
    hw_addr 00:00:00:00:00:00 state inactive opstate detached roce enable
[port,new] pci/0000:08:00.0/32768: type notset flavour pcisf controller 0 pfnum 0 sfnum 107 splittable false
  function:
    hw_addr 00:00:00:00:00:00 state inactive opstate detached roce enable
[port,del] pci/0000:08:00.0/32768: type notset flavour pcisf controller 0 pfnum 0 sfnum 107 splittable false
  function:
    hw_addr 00:00:00:00:00:00 state inactive opstate detached roce enable

---
v4->v5:
- converted priv pointer in netlink_sock to genl_sock container,
  containing xarray pointer
- introduced per-family init/destroy callbacks and priv_size to allocate
  per-sock private, converted devlink to that
- see patches #5 and #8 for more details
v3->v4:
- converted from sk_user_data pointer use to nlk(sk)->priv pointer and
  allow priv to be stored for multiple generic netlink families, see
  patch #5 for more details
v2->v3:
- small cosmetical fixes in patch #6
v1->v2:
- added patch #6, fixed generated docs
- see individual patches for details

Jiri Pirko (9):
  devlink: use devl_is_registered() helper instead xa_get_mark()
  devlink: introduce __devl_is_registered() helper and use it instead of
    xa_get_mark()
  devlink: send notifications only if there are listeners
  devlink: introduce a helper for netlink multicast send
  genetlink: introduce per-sock family private storage
  netlink: introduce typedef for filter function
  genetlink: introduce helpers to do filtered multicast
  devlink: add a command to set notification filter and use it for
    multicasts
  devlink: extend multicast filtering by port index

 Documentation/netlink/specs/devlink.yaml |  11 ++
 drivers/connector/connector.c            |   5 +-
 include/linux/connector.h                |   3 +-
 include/linux/netlink.h                  |   6 +-
 include/net/genetlink.h                  |  41 ++++++-
 include/net/netlink.h                    |  31 ++++-
 include/uapi/linux/devlink.h             |   2 +
 net/devlink/dev.c                        |  13 +-
 net/devlink/devl_internal.h              |  59 ++++++++-
 net/devlink/health.c                     |  10 +-
 net/devlink/linecard.c                   |   5 +-
 net/devlink/netlink.c                    | 116 ++++++++++++++++++
 net/devlink/netlink_gen.c                |  16 ++-
 net/devlink/netlink_gen.h                |   4 +-
 net/devlink/param.c                      |   5 +-
 net/devlink/port.c                       |   8 +-
 net/devlink/rate.c                       |   5 +-
 net/devlink/region.c                     |   6 +-
 net/devlink/trap.c                       |  18 +--
 net/netlink/af_netlink.c                 |   5 +-
 net/netlink/af_netlink.h                 |  15 +++
 net/netlink/genetlink.c                  | 146 +++++++++++++++++++++++
 22 files changed, 477 insertions(+), 53 deletions(-)

-- 
2.41.0

[patch net-next v5 1/9] devlink: use devl_is_registered() helper instead xa_get_mark()

[Jiri Pirko]

From: Jiri Pirko <jiri@nvidia.com>

Instead of checking the xarray mark directly using xa_get_mark() helper
use devl_is_registered() helper which wraps it up. Note that there are
couple more users of xa_get_mark() left which are going to be handled
by the next patch.

Signed-off-by: Jiri Pirko <jiri@nvidia.com>
---
 net/devlink/dev.c  | 4 ++--
 net/devlink/rate.c | 2 +-
 net/devlink/trap.c | 9 ++++++---
 3 files changed, 9 insertions(+), 6 deletions(-)

diff --git a/net/devlink/dev.c b/net/devlink/dev.c
index 918a0395b03e..3fe93c8a9fe2 100644
--- a/net/devlink/dev.c
+++ b/net/devlink/dev.c
@@ -202,7 +202,7 @@ static void devlink_notify(struct devlink *devlink, enum devlink_command cmd)
 	int err;
 
 	WARN_ON(cmd != DEVLINK_CMD_NEW && cmd != DEVLINK_CMD_DEL);
-	WARN_ON(!xa_get_mark(&devlinks, devlink->index, DEVLINK_REGISTERED));
+	WARN_ON(!devl_is_registered(devlink));
 
 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
 	if (!msg)
@@ -999,7 +999,7 @@ static void __devlink_flash_update_notify(struct devlink *devlink,
 		cmd != DEVLINK_CMD_FLASH_UPDATE_END &&
 		cmd != DEVLINK_CMD_FLASH_UPDATE_STATUS);
 
-	if (!xa_get_mark(&devlinks, devlink->index, DEVLINK_REGISTERED))
+	if (!devl_is_registered(devlink))
 		return;
 
 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
diff --git a/net/devlink/rate.c b/net/devlink/rate.c
index 94b289b93ff2..e2190cf22beb 100644
--- a/net/devlink/rate.c
+++ b/net/devlink/rate.c
@@ -146,7 +146,7 @@ static void devlink_rate_notify(struct devlink_rate *devlink_rate,
 
 	WARN_ON(cmd != DEVLINK_CMD_RATE_NEW && cmd != DEVLINK_CMD_RATE_DEL);
 
-	if (!xa_get_mark(&devlinks, devlink->index, DEVLINK_REGISTERED))
+	if (!devl_is_registered(devlink))
 		return;
 
 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
diff --git a/net/devlink/trap.c b/net/devlink/trap.c
index c26313e7ca08..908085e2c990 100644
--- a/net/devlink/trap.c
+++ b/net/devlink/trap.c
@@ -1173,7 +1173,8 @@ devlink_trap_group_notify(struct devlink *devlink,
 
 	WARN_ON_ONCE(cmd != DEVLINK_CMD_TRAP_GROUP_NEW &&
 		     cmd != DEVLINK_CMD_TRAP_GROUP_DEL);
-	if (!xa_get_mark(&devlinks, devlink->index, DEVLINK_REGISTERED))
+
+	if (!devl_is_registered(devlink))
 		return;
 
 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
@@ -1234,7 +1235,8 @@ static void devlink_trap_notify(struct devlink *devlink,
 
 	WARN_ON_ONCE(cmd != DEVLINK_CMD_TRAP_NEW &&
 		     cmd != DEVLINK_CMD_TRAP_DEL);
-	if (!xa_get_mark(&devlinks, devlink->index, DEVLINK_REGISTERED))
+
+	if (!devl_is_registered(devlink))
 		return;
 
 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
@@ -1710,7 +1712,8 @@ devlink_trap_policer_notify(struct devlink *devlink,
 
 	WARN_ON_ONCE(cmd != DEVLINK_CMD_TRAP_POLICER_NEW &&
 		     cmd != DEVLINK_CMD_TRAP_POLICER_DEL);
-	if (!xa_get_mark(&devlinks, devlink->index, DEVLINK_REGISTERED))
+
+	if (!devl_is_registered(devlink))
 		return;
 
 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
-- 
2.41.0

[patch net-next v5 2/9] devlink: introduce __devl_is_registered() helper and use it instead of xa_get_mark()

[Jiri Pirko]

From: Jiri Pirko <jiri@nvidia.com>

Introduce __devl_is_registered() which does not assert on devlink
instance lock and use it in notifications which may be called
without devlink instance lock held.

Signed-off-by: Jiri Pirko <jiri@nvidia.com>
---
 net/devlink/devl_internal.h | 7 ++++++-
 net/devlink/linecard.c      | 2 +-
 net/devlink/port.c          | 2 +-
 net/devlink/region.c        | 3 ++-
 4 files changed, 10 insertions(+), 4 deletions(-)

diff --git a/net/devlink/devl_internal.h b/net/devlink/devl_internal.h
index 5ea2e2012e93..59ae4761d10a 100644
--- a/net/devlink/devl_internal.h
+++ b/net/devlink/devl_internal.h
@@ -91,10 +91,15 @@ extern struct genl_family devlink_nl_family;
 
 struct devlink *devlinks_xa_find_get(struct net *net, unsigned long *indexp);
 
+static inline bool __devl_is_registered(struct devlink *devlink)
+{
+	return xa_get_mark(&devlinks, devlink->index, DEVLINK_REGISTERED);
+}
+
 static inline bool devl_is_registered(struct devlink *devlink)
 {
 	devl_assert_locked(devlink);
-	return xa_get_mark(&devlinks, devlink->index, DEVLINK_REGISTERED);
+	return __devl_is_registered(devlink);
 }
 
 static inline void devl_dev_lock(struct devlink *devlink, bool dev_lock)
diff --git a/net/devlink/linecard.c b/net/devlink/linecard.c
index 2f1c317b64cd..9d080ac1734b 100644
--- a/net/devlink/linecard.c
+++ b/net/devlink/linecard.c
@@ -136,7 +136,7 @@ static void devlink_linecard_notify(struct devlink_linecard *linecard,
 	WARN_ON(cmd != DEVLINK_CMD_LINECARD_NEW &&
 		cmd != DEVLINK_CMD_LINECARD_DEL);
 
-	if (!xa_get_mark(&devlinks, devlink->index, DEVLINK_REGISTERED))
+	if (!__devl_is_registered(devlink))
 		return;
 
 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
diff --git a/net/devlink/port.c b/net/devlink/port.c
index 7634f187fa50..f229a8699214 100644
--- a/net/devlink/port.c
+++ b/net/devlink/port.c
@@ -512,7 +512,7 @@ static void devlink_port_notify(struct devlink_port *devlink_port,
 
 	WARN_ON(cmd != DEVLINK_CMD_PORT_NEW && cmd != DEVLINK_CMD_PORT_DEL);
 
-	if (!xa_get_mark(&devlinks, devlink->index, DEVLINK_REGISTERED))
+	if (!__devl_is_registered(devlink))
 		return;
 
 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
diff --git a/net/devlink/region.c b/net/devlink/region.c
index e3bab458db94..b65181aa269a 100644
--- a/net/devlink/region.c
+++ b/net/devlink/region.c
@@ -234,7 +234,8 @@ static void devlink_nl_region_notify(struct devlink_region *region,
 	struct sk_buff *msg;
 
 	WARN_ON(cmd != DEVLINK_CMD_REGION_NEW && cmd != DEVLINK_CMD_REGION_DEL);
-	if (!xa_get_mark(&devlinks, devlink->index, DEVLINK_REGISTERED))
+
+	if (!__devl_is_registered(devlink))
 		return;
 
 	msg = devlink_nl_region_notify_build(region, snapshot, cmd, 0, 0);
-- 
2.41.0

[patch net-next v5 3/9] devlink: send notifications only if there are listeners

[Jiri Pirko]

From: Jiri Pirko <jiri@nvidia.com>

Introduce devlink_nl_notify_need() helper and using it to check at the
beginning of notification functions to avoid overhead of composing
notification messages in case nobody listens.

Signed-off-by: Jiri Pirko <jiri@nvidia.com>
---
v4->v5:
- swapped order in test in param.c
---
 net/devlink/dev.c           | 5 ++++-
 net/devlink/devl_internal.h | 6 ++++++
 net/devlink/health.c        | 3 +++
 net/devlink/linecard.c      | 2 +-
 net/devlink/param.c         | 2 +-
 net/devlink/port.c          | 2 +-
 net/devlink/rate.c          | 2 +-
 net/devlink/region.c        | 2 +-
 net/devlink/trap.c          | 6 +++---
 9 files changed, 21 insertions(+), 9 deletions(-)

diff --git a/net/devlink/dev.c b/net/devlink/dev.c
index 3fe93c8a9fe2..63fe3e02c928 100644
--- a/net/devlink/dev.c
+++ b/net/devlink/dev.c
@@ -204,6 +204,9 @@ static void devlink_notify(struct devlink *devlink, enum devlink_command cmd)
 	WARN_ON(cmd != DEVLINK_CMD_NEW && cmd != DEVLINK_CMD_DEL);
 	WARN_ON(!devl_is_registered(devlink));
 
+	if (!devlink_nl_notify_need(devlink))
+		return;
+
 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
 	if (!msg)
 		return;
@@ -999,7 +1002,7 @@ static void __devlink_flash_update_notify(struct devlink *devlink,
 		cmd != DEVLINK_CMD_FLASH_UPDATE_END &&
 		cmd != DEVLINK_CMD_FLASH_UPDATE_STATUS);
 
-	if (!devl_is_registered(devlink))
+	if (!devl_is_registered(devlink) || !devlink_nl_notify_need(devlink))
 		return;
 
 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
diff --git a/net/devlink/devl_internal.h b/net/devlink/devl_internal.h
index 59ae4761d10a..510990de094e 100644
--- a/net/devlink/devl_internal.h
+++ b/net/devlink/devl_internal.h
@@ -185,6 +185,12 @@ int devlink_nl_put_nested_handle(struct sk_buff *msg, struct net *net,
 				 struct devlink *devlink, int attrtype);
 int devlink_nl_msg_reply_and_new(struct sk_buff **msg, struct genl_info *info);
 
+static inline bool devlink_nl_notify_need(struct devlink *devlink)
+{
+	return genl_has_listeners(&devlink_nl_family, devlink_net(devlink),
+				  DEVLINK_MCGRP_CONFIG);
+}
+
 /* Notify */
 void devlink_notify_register(struct devlink *devlink);
 void devlink_notify_unregister(struct devlink *devlink);
diff --git a/net/devlink/health.c b/net/devlink/health.c
index 71ae121dc739..0795dcf22ca8 100644
--- a/net/devlink/health.c
+++ b/net/devlink/health.c
@@ -496,6 +496,9 @@ static void devlink_recover_notify(struct devlink_health_reporter *reporter,
 	WARN_ON(cmd != DEVLINK_CMD_HEALTH_REPORTER_RECOVER);
 	ASSERT_DEVLINK_REGISTERED(devlink);
 
+	if (!devlink_nl_notify_need(devlink))
+		return;
+
 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
 	if (!msg)
 		return;
diff --git a/net/devlink/linecard.c b/net/devlink/linecard.c
index 9d080ac1734b..45b36975ee6f 100644
--- a/net/devlink/linecard.c
+++ b/net/devlink/linecard.c
@@ -136,7 +136,7 @@ static void devlink_linecard_notify(struct devlink_linecard *linecard,
 	WARN_ON(cmd != DEVLINK_CMD_LINECARD_NEW &&
 		cmd != DEVLINK_CMD_LINECARD_DEL);
 
-	if (!__devl_is_registered(devlink))
+	if (!__devl_is_registered(devlink) || !devlink_nl_notify_need(devlink))
 		return;
 
 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
diff --git a/net/devlink/param.c b/net/devlink/param.c
index d74df09311a9..7516b524ffb7 100644
--- a/net/devlink/param.c
+++ b/net/devlink/param.c
@@ -343,7 +343,7 @@ static void devlink_param_notify(struct devlink *devlink,
 	 * will replay the notifications if the params are added/removed
 	 * outside of the lifetime of the instance.
 	 */
-	if (!devl_is_registered(devlink))
+	if (!devl_is_registered(devlink) || !devlink_nl_notify_need(devlink))
 		return;
 
 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
diff --git a/net/devlink/port.c b/net/devlink/port.c
index f229a8699214..32f4d0331e63 100644
--- a/net/devlink/port.c
+++ b/net/devlink/port.c
@@ -512,7 +512,7 @@ static void devlink_port_notify(struct devlink_port *devlink_port,
 
 	WARN_ON(cmd != DEVLINK_CMD_PORT_NEW && cmd != DEVLINK_CMD_PORT_DEL);
 
-	if (!__devl_is_registered(devlink))
+	if (!__devl_is_registered(devlink) || !devlink_nl_notify_need(devlink))
 		return;
 
 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
diff --git a/net/devlink/rate.c b/net/devlink/rate.c
index e2190cf22beb..0371a2dd3e0a 100644
--- a/net/devlink/rate.c
+++ b/net/devlink/rate.c
@@ -146,7 +146,7 @@ static void devlink_rate_notify(struct devlink_rate *devlink_rate,
 
 	WARN_ON(cmd != DEVLINK_CMD_RATE_NEW && cmd != DEVLINK_CMD_RATE_DEL);
 
-	if (!devl_is_registered(devlink))
+	if (!devl_is_registered(devlink) || !devlink_nl_notify_need(devlink))
 		return;
 
 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
diff --git a/net/devlink/region.c b/net/devlink/region.c
index b65181aa269a..bf61312f64bd 100644
--- a/net/devlink/region.c
+++ b/net/devlink/region.c
@@ -235,7 +235,7 @@ static void devlink_nl_region_notify(struct devlink_region *region,
 
 	WARN_ON(cmd != DEVLINK_CMD_REGION_NEW && cmd != DEVLINK_CMD_REGION_DEL);
 
-	if (!__devl_is_registered(devlink))
+	if (!__devl_is_registered(devlink) || !devlink_nl_notify_need(devlink))
 		return;
 
 	msg = devlink_nl_region_notify_build(region, snapshot, cmd, 0, 0);
diff --git a/net/devlink/trap.c b/net/devlink/trap.c
index 908085e2c990..3ca1ca7e2e64 100644
--- a/net/devlink/trap.c
+++ b/net/devlink/trap.c
@@ -1174,7 +1174,7 @@ devlink_trap_group_notify(struct devlink *devlink,
 	WARN_ON_ONCE(cmd != DEVLINK_CMD_TRAP_GROUP_NEW &&
 		     cmd != DEVLINK_CMD_TRAP_GROUP_DEL);
 
-	if (!devl_is_registered(devlink))
+	if (!devl_is_registered(devlink) || !devlink_nl_notify_need(devlink))
 		return;
 
 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
@@ -1236,7 +1236,7 @@ static void devlink_trap_notify(struct devlink *devlink,
 	WARN_ON_ONCE(cmd != DEVLINK_CMD_TRAP_NEW &&
 		     cmd != DEVLINK_CMD_TRAP_DEL);
 
-	if (!devl_is_registered(devlink))
+	if (!devl_is_registered(devlink) || !devlink_nl_notify_need(devlink))
 		return;
 
 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
@@ -1713,7 +1713,7 @@ devlink_trap_policer_notify(struct devlink *devlink,
 	WARN_ON_ONCE(cmd != DEVLINK_CMD_TRAP_POLICER_NEW &&
 		     cmd != DEVLINK_CMD_TRAP_POLICER_DEL);
 
-	if (!devl_is_registered(devlink))
+	if (!devl_is_registered(devlink) || !devlink_nl_notify_need(devlink))
 		return;
 
 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
-- 
2.41.0

[patch net-next v5 4/9] devlink: introduce a helper for netlink multicast send

[Jiri Pirko]

From: Jiri Pirko <jiri@nvidia.com>

Introduce a helper devlink_nl_notify_send() so each object notification
function does not have to call genlmsg_multicast_netns() with the same
arguments.

Signed-off-by: Jiri Pirko <jiri@nvidia.com>
---
 net/devlink/dev.c           | 6 ++----
 net/devlink/devl_internal.h | 7 +++++++
 net/devlink/health.c        | 3 +--
 net/devlink/linecard.c      | 3 +--
 net/devlink/param.c         | 3 +--
 net/devlink/port.c          | 3 +--
 net/devlink/rate.c          | 3 +--
 net/devlink/region.c        | 3 +--
 net/devlink/trap.c          | 9 +++------
 9 files changed, 18 insertions(+), 22 deletions(-)

diff --git a/net/devlink/dev.c b/net/devlink/dev.c
index 63fe3e02c928..19dbf540748a 100644
--- a/net/devlink/dev.c
+++ b/net/devlink/dev.c
@@ -217,8 +217,7 @@ static void devlink_notify(struct devlink *devlink, enum devlink_command cmd)
 		return;
 	}
 
-	genlmsg_multicast_netns(&devlink_nl_family, devlink_net(devlink),
-				msg, 0, DEVLINK_MCGRP_CONFIG, GFP_KERNEL);
+	devlink_nl_notify_send(devlink, msg);
 }
 
 int devlink_nl_get_doit(struct sk_buff *skb, struct genl_info *info)
@@ -1013,8 +1012,7 @@ static void __devlink_flash_update_notify(struct devlink *devlink,
 	if (err)
 		goto out_free_msg;
 
-	genlmsg_multicast_netns(&devlink_nl_family, devlink_net(devlink),
-				msg, 0, DEVLINK_MCGRP_CONFIG, GFP_KERNEL);
+	devlink_nl_notify_send(devlink, msg);
 	return;
 
 out_free_msg:
diff --git a/net/devlink/devl_internal.h b/net/devlink/devl_internal.h
index 510990de094e..84dc9628d3f2 100644
--- a/net/devlink/devl_internal.h
+++ b/net/devlink/devl_internal.h
@@ -191,6 +191,13 @@ static inline bool devlink_nl_notify_need(struct devlink *devlink)
 				  DEVLINK_MCGRP_CONFIG);
 }
 
+static inline void devlink_nl_notify_send(struct devlink *devlink,
+					  struct sk_buff *msg)
+{
+	genlmsg_multicast_netns(&devlink_nl_family, devlink_net(devlink),
+				msg, 0, DEVLINK_MCGRP_CONFIG, GFP_KERNEL);
+}
+
 /* Notify */
 void devlink_notify_register(struct devlink *devlink);
 void devlink_notify_unregister(struct devlink *devlink);
diff --git a/net/devlink/health.c b/net/devlink/health.c
index 0795dcf22ca8..1d59ec0202f6 100644
--- a/net/devlink/health.c
+++ b/net/devlink/health.c
@@ -509,8 +509,7 @@ static void devlink_recover_notify(struct devlink_health_reporter *reporter,
 		return;
 	}
 
-	genlmsg_multicast_netns(&devlink_nl_family, devlink_net(devlink), msg,
-				0, DEVLINK_MCGRP_CONFIG, GFP_KERNEL);
+	devlink_nl_notify_send(devlink, msg);
 }
 
 void
diff --git a/net/devlink/linecard.c b/net/devlink/linecard.c
index 45b36975ee6f..67f70a621d27 100644
--- a/net/devlink/linecard.c
+++ b/net/devlink/linecard.c
@@ -150,8 +150,7 @@ static void devlink_linecard_notify(struct devlink_linecard *linecard,
 		return;
 	}
 
-	genlmsg_multicast_netns(&devlink_nl_family, devlink_net(devlink),
-				msg, 0, DEVLINK_MCGRP_CONFIG, GFP_KERNEL);
+	devlink_nl_notify_send(devlink, msg);
 }
 
 void devlink_linecards_notify_register(struct devlink *devlink)
diff --git a/net/devlink/param.c b/net/devlink/param.c
index 7516b524ffb7..22bc3b500518 100644
--- a/net/devlink/param.c
+++ b/net/devlink/param.c
@@ -356,8 +356,7 @@ static void devlink_param_notify(struct devlink *devlink,
 		return;
 	}
 
-	genlmsg_multicast_netns(&devlink_nl_family, devlink_net(devlink),
-				msg, 0, DEVLINK_MCGRP_CONFIG, GFP_KERNEL);
+	devlink_nl_notify_send(devlink, msg);
 }
 
 static void devlink_params_notify(struct devlink *devlink,
diff --git a/net/devlink/port.c b/net/devlink/port.c
index 32f4d0331e63..758df3000a1b 100644
--- a/net/devlink/port.c
+++ b/net/devlink/port.c
@@ -525,8 +525,7 @@ static void devlink_port_notify(struct devlink_port *devlink_port,
 		return;
 	}
 
-	genlmsg_multicast_netns(&devlink_nl_family, devlink_net(devlink), msg,
-				0, DEVLINK_MCGRP_CONFIG, GFP_KERNEL);
+	devlink_nl_notify_send(devlink, msg);
 }
 
 static void devlink_ports_notify(struct devlink *devlink,
diff --git a/net/devlink/rate.c b/net/devlink/rate.c
index 0371a2dd3e0a..7139e67e93ae 100644
--- a/net/devlink/rate.c
+++ b/net/devlink/rate.c
@@ -159,8 +159,7 @@ static void devlink_rate_notify(struct devlink_rate *devlink_rate,
 		return;
 	}
 
-	genlmsg_multicast_netns(&devlink_nl_family, devlink_net(devlink), msg,
-				0, DEVLINK_MCGRP_CONFIG, GFP_KERNEL);
+	devlink_nl_notify_send(devlink, msg);
 }
 
 void devlink_rates_notify_register(struct devlink *devlink)
diff --git a/net/devlink/region.c b/net/devlink/region.c
index bf61312f64bd..7319127c5913 100644
--- a/net/devlink/region.c
+++ b/net/devlink/region.c
@@ -242,8 +242,7 @@ static void devlink_nl_region_notify(struct devlink_region *region,
 	if (IS_ERR(msg))
 		return;
 
-	genlmsg_multicast_netns(&devlink_nl_family, devlink_net(devlink), msg,
-				0, DEVLINK_MCGRP_CONFIG, GFP_KERNEL);
+	devlink_nl_notify_send(devlink, msg);
 }
 
 void devlink_regions_notify_register(struct devlink *devlink)
diff --git a/net/devlink/trap.c b/net/devlink/trap.c
index 3ca1ca7e2e64..5d18c7424df1 100644
--- a/net/devlink/trap.c
+++ b/net/devlink/trap.c
@@ -1188,8 +1188,7 @@ devlink_trap_group_notify(struct devlink *devlink,
 		return;
 	}
 
-	genlmsg_multicast_netns(&devlink_nl_family, devlink_net(devlink),
-				msg, 0, DEVLINK_MCGRP_CONFIG, GFP_KERNEL);
+	devlink_nl_notify_send(devlink, msg);
 }
 
 void devlink_trap_groups_notify_register(struct devlink *devlink)
@@ -1249,8 +1248,7 @@ static void devlink_trap_notify(struct devlink *devlink,
 		return;
 	}
 
-	genlmsg_multicast_netns(&devlink_nl_family, devlink_net(devlink),
-				msg, 0, DEVLINK_MCGRP_CONFIG, GFP_KERNEL);
+	devlink_nl_notify_send(devlink, msg);
 }
 
 void devlink_traps_notify_register(struct devlink *devlink)
@@ -1727,8 +1725,7 @@ devlink_trap_policer_notify(struct devlink *devlink,
 		return;
 	}
 
-	genlmsg_multicast_netns(&devlink_nl_family, devlink_net(devlink),
-				msg, 0, DEVLINK_MCGRP_CONFIG, GFP_KERNEL);
+	devlink_nl_notify_send(devlink, msg);
 }
 
 void devlink_trap_policers_notify_register(struct devlink *devlink)
-- 
2.41.0

[patch net-next v5 5/9] genetlink: introduce per-sock family private storage

[Jiri Pirko]

From: Jiri Pirko <jiri@nvidia.com>

Introduce a wrapper sock struct for Generic netlink and store
a pointer to family privs xarray. This per socket xarray contains
family->id indexed priv storage.

Note I used xarray instead of suggested linked list as it is more
convenient.

Introduce genl_sk_priv_get() to get the family priv pointer and
initialize it in case it does not exist.
Introduce __genl_sk_priv_get() to obtain family the priv pointer
under RCU read lock.

Allow family to specify the priv size, init() and destroy() callbacks.

Signed-off-by: Jiri Pirko <jiri@nvidia.com>
---
v4->v5:
- s/Returns/Return/ in function comments
- introduced wrapper genl sock struct and store xarray there
- changed family helpers to genl_sk_priv_get() and __genl_sk_priv_get()
- introduced sock_priv_size for family and use this to allocate the priv
  in generic netlink code
- introduced init/destroy callbacks for family privs
v3->v4:
- new patch
---
 include/net/genetlink.h  |   6 ++
 net/netlink/af_netlink.c |   2 +-
 net/netlink/af_netlink.h |  15 ++++
 net/netlink/genetlink.c  | 146 +++++++++++++++++++++++++++++++++++++++
 4 files changed, 168 insertions(+), 1 deletion(-)

diff --git a/include/net/genetlink.h b/include/net/genetlink.h
index e18a4c0d69ee..dbf11464e96a 100644
--- a/include/net/genetlink.h
+++ b/include/net/genetlink.h
@@ -87,6 +87,9 @@ struct genl_family {
 	int			id;
 	/* starting number of multicast group IDs in this family */
 	unsigned int		mcgrp_offset;
+	size_t			sock_priv_size;
+	void			(*sock_priv_init)(void *priv);
+	void			(*sock_priv_destroy)(void *priv);
 };
 
 /**
@@ -301,6 +304,9 @@ int genl_unregister_family(const struct genl_family *family);
 void genl_notify(const struct genl_family *family, struct sk_buff *skb,
 		 struct genl_info *info, u32 group, gfp_t flags);
 
+void *__genl_sk_priv_get(struct sock *sk, struct genl_family *family);
+void *genl_sk_priv_get(struct sock *sk, struct genl_family *family);
+
 void *genlmsg_put(struct sk_buff *skb, u32 portid, u32 seq,
 		  const struct genl_family *family, int flags, u8 cmd);
 
diff --git a/net/netlink/af_netlink.c b/net/netlink/af_netlink.c
index 177126fb0484..5683b0ca23b1 100644
--- a/net/netlink/af_netlink.c
+++ b/net/netlink/af_netlink.c
@@ -632,7 +632,7 @@ static void netlink_remove(struct sock *sk)
 static struct proto netlink_proto = {
 	.name	  = "NETLINK",
 	.owner	  = THIS_MODULE,
-	.obj_size = sizeof(struct netlink_sock),
+	.obj_size = NETLINK_SOCK_SIZE,
 };
 
 static int __netlink_create(struct net *net, struct socket *sock,
diff --git a/net/netlink/af_netlink.h b/net/netlink/af_netlink.h
index 2145979b9986..1b3ed8919574 100644
--- a/net/netlink/af_netlink.h
+++ b/net/netlink/af_netlink.h
@@ -60,6 +60,21 @@ static inline struct netlink_sock *nlk_sk(struct sock *sk)
 
 #define nlk_test_bit(nr, sk) test_bit(NETLINK_F_##nr, &nlk_sk(sk)->flags)
 
+struct genl_sock {
+	struct netlink_sock nlk_sk;
+	struct xarray *family_privs;
+};
+
+static inline struct genl_sock *genl_sk(struct sock *sk)
+{
+	return container_of(nlk_sk(sk), struct genl_sock, nlk_sk);
+}
+
+/* Size of netlink sock is size of the biggest user with priv,
+ * which is currently just Generic Netlink.
+ */
+#define NETLINK_SOCK_SIZE sizeof(struct genl_sock)
+
 struct netlink_table {
 	struct rhashtable	hash;
 	struct hlist_head	mc_list;
diff --git a/net/netlink/genetlink.c b/net/netlink/genetlink.c
index 92ef5ed2e7b0..51720c2c6bda 100644
--- a/net/netlink/genetlink.c
+++ b/net/netlink/genetlink.c
@@ -22,6 +22,8 @@
 #include <net/sock.h>
 #include <net/genetlink.h>
 
+#include "af_netlink.h"
+
 static DEFINE_MUTEX(genl_mutex); /* serialization of message processing */
 static DECLARE_RWSEM(cb_lock);
 
@@ -1699,12 +1701,156 @@ static int genl_bind(struct net *net, int group)
 	return ret;
 }
 
+struct genl_sk_priv {
+	void (*destructor)(void *priv);
+	long priv[];
+};
+
+static struct genl_sk_priv *genl_sk_priv_alloc(struct genl_family *family)
+{
+	struct genl_sk_priv *priv;
+
+	priv = kzalloc(size_add(sizeof(*priv), family->sock_priv_size),
+		       GFP_KERNEL);
+	if (!priv)
+		return ERR_PTR(-ENOMEM);
+	priv->destructor = family->sock_priv_destroy;
+	if (family->sock_priv_init)
+		family->sock_priv_init(priv->priv);
+	return priv;
+}
+
+static void genl_sk_priv_free(struct genl_sk_priv *priv)
+{
+	if (priv->destructor)
+		priv->destructor(priv->priv);
+	kfree(priv);
+}
+
+static void genl_release(struct sock *sk, unsigned long *groups)
+{
+	struct genl_sock *gsk = genl_sk(sk);
+	struct genl_sk_priv *priv;
+	unsigned long family_id;
+
+	if (!gsk->family_privs)
+		return;
+	xa_for_each(gsk->family_privs, family_id, priv) {
+		xa_erase(gsk->family_privs, family_id);
+		genl_sk_priv_free(priv);
+	}
+	xa_destroy(gsk->family_privs);
+	kfree(gsk->family_privs);
+}
+
+static struct xarray *genl_family_privs_get(struct genl_sock *gsk)
+{
+	struct xarray *family_privs;
+
+again:
+	family_privs = READ_ONCE(gsk->family_privs);
+	if (family_privs)
+		return family_privs;
+
+	family_privs = kzalloc(sizeof(*family_privs), GFP_KERNEL);
+	if (!family_privs)
+		return ERR_PTR(-ENOMEM);
+	xa_init_flags(family_privs, XA_FLAGS_ALLOC);
+
+	/* Use genl lock to protect family_privs to be
+	 * initialized in parallel by different CPU.
+	 */
+	genl_lock();
+	if (unlikely(gsk->family_privs)) {
+		xa_destroy(family_privs);
+		kfree(family_privs);
+		genl_unlock();
+		goto again;
+	}
+	WRITE_ONCE(gsk->family_privs, family_privs);
+	genl_unlock();
+	return family_privs;
+}
+
+/**
+ * __genl_sk_priv_get - Get per-socket private pointer for family
+ *
+ * @sk: socket
+ * @family: family
+ *
+ * Lookup a private pointer stored per-socket by a specified
+ * Generic netlink family.
+ *
+ * Caller should make sure this is called in RCU read locked section.
+ *
+ * Return: valid pointer on success, otherwise NULL.
+ */
+void *__genl_sk_priv_get(struct sock *sk, struct genl_family *family)
+{
+	struct genl_sock *gsk = genl_sk(sk);
+	struct genl_sk_priv *priv;
+	struct xarray *family_privs;
+
+	family_privs = READ_ONCE(gsk->family_privs);
+	if (!family_privs)
+		return NULL;
+	priv = xa_load(family_privs, family->id);
+	return priv ? priv->priv : NULL;
+}
+
+/**
+ * genl_sk_priv_get - Get per-socket private pointer for family
+ *
+ * @sk: socket
+ * @family: family
+ *
+ * Store a private pointer per-socket for a specified
+ * Generic netlink family.
+ *
+ * Caller has to make sure this is not called in parallel multiple times
+ * for the same sock and also in parallel to genl_release() for the same sock.
+ *
+ * Return: previously stored private pointer for the family (could be NULL)
+ * on success, otherwise negative error value encoded by ERR_PTR().
+ */
+void *genl_sk_priv_get(struct sock *sk, struct genl_family *family)
+{
+	struct genl_sk_priv *priv, *old_priv;
+	struct genl_sock *gsk = genl_sk(sk);
+	struct xarray *family_privs;
+
+	family_privs = genl_family_privs_get(gsk);
+	if (IS_ERR(family_privs))
+		return ERR_CAST(family_privs);
+
+	priv = xa_load(family_privs, family->id);
+	if (priv)
+		return priv->priv;
+
+	/* priv for the family does not exist so far, create it. */
+
+	priv = genl_sk_priv_alloc(family);
+	if (IS_ERR(priv))
+		return ERR_CAST(priv);
+
+	old_priv = xa_cmpxchg(family_privs, family->id, NULL, priv, GFP_KERNEL);
+	if (xa_is_err(old_priv))
+		return ERR_PTR(xa_err(old_priv));
+	else if (!old_priv)
+		return priv->priv;
+
+	/* Race happened, priv was already inserted. */
+	genl_sk_priv_free(priv);
+	return old_priv->priv;
+}
+
 static int __net_init genl_pernet_init(struct net *net)
 {
 	struct netlink_kernel_cfg cfg = {
 		.input		= genl_rcv,
 		.flags		= NL_CFG_F_NONROOT_RECV,
 		.bind		= genl_bind,
+		.release	= genl_release,
 	};
 
 	/* we'll bump the group number right afterwards */
-- 
2.41.0

[patch net-next v5 6/9] netlink: introduce typedef for filter function

[Jiri Pirko]

From: Jiri Pirko <jiri@nvidia.com>

Make the code using filter function a bit nicer by consolidating the
filter function arguments using typedef.

Suggested-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Jiri Pirko <jiri@nvidia.com>
---
v2->v3:
- left the original .c and .h arg names and types
  inconsistencies for cn_netlink_send_mult() and
  netlink_broadcast_filtered()
v1->v2:
- new patch
---
 drivers/connector/connector.c | 5 ++---
 include/linux/connector.h     | 3 +--
 include/linux/netlink.h       | 6 ++++--
 net/netlink/af_netlink.c      | 3 +--
 4 files changed, 8 insertions(+), 9 deletions(-)

diff --git a/drivers/connector/connector.c b/drivers/connector/connector.c
index 7f7b94f616a6..4028e8eeba82 100644
--- a/drivers/connector/connector.c
+++ b/drivers/connector/connector.c
@@ -59,9 +59,8 @@ static int cn_already_initialized;
  * both, or if both are zero then the group is looked up and sent there.
  */
 int cn_netlink_send_mult(struct cn_msg *msg, u16 len, u32 portid, u32 __group,
-	gfp_t gfp_mask,
-	int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
-	void *filter_data)
+			 gfp_t gfp_mask, netlink_filter_fn filter,
+			 void *filter_data)
 {
 	struct cn_callback_entry *__cbq;
 	unsigned int size;
diff --git a/include/linux/connector.h b/include/linux/connector.h
index cec2d99ae902..70bc1160f3d8 100644
--- a/include/linux/connector.h
+++ b/include/linux/connector.h
@@ -100,8 +100,7 @@ void cn_del_callback(const struct cb_id *id);
  */
 int cn_netlink_send_mult(struct cn_msg *msg, u16 len, u32 portid,
 			 u32 group, gfp_t gfp_mask,
-			 int (*filter)(struct sock *dsk, struct sk_buff *skb,
-				       void *data),
+			 netlink_filter_fn filter,
 			 void *filter_data);
 
 /**
diff --git a/include/linux/netlink.h b/include/linux/netlink.h
index abe91ed6b9aa..1a4445bf2ab9 100644
--- a/include/linux/netlink.h
+++ b/include/linux/netlink.h
@@ -228,10 +228,12 @@ bool netlink_strict_get_check(struct sk_buff *skb);
 int netlink_unicast(struct sock *ssk, struct sk_buff *skb, __u32 portid, int nonblock);
 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, __u32 portid,
 		      __u32 group, gfp_t allocation);
+
+typedef int (*netlink_filter_fn)(struct sock *dsk, struct sk_buff *skb, void *data);
+
 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb,
 			       __u32 portid, __u32 group, gfp_t allocation,
-			       int (*filter)(struct sock *dsk,
-					     struct sk_buff *skb, void *data),
+			       netlink_filter_fn filter,
 			       void *filter_data);
 int netlink_set_err(struct sock *ssk, __u32 portid, __u32 group, int code);
 int netlink_register_notifier(struct notifier_block *nb);
diff --git a/net/netlink/af_netlink.c b/net/netlink/af_netlink.c
index 5683b0ca23b1..0efc1bd451fc 100644
--- a/net/netlink/af_netlink.c
+++ b/net/netlink/af_netlink.c
@@ -1519,8 +1519,7 @@ static void do_one_broadcast(struct sock *sk,
 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb,
 			       u32 portid,
 			       u32 group, gfp_t allocation,
-			       int (*filter)(struct sock *dsk,
-					     struct sk_buff *skb, void *data),
+			       netlink_filter_fn filter,
 			       void *filter_data)
 {
 	struct net *net = sock_net(ssk);
-- 
2.41.0

[patch net-next v5 7/9] genetlink: introduce helpers to do filtered multicast

[Jiri Pirko]

From: Jiri Pirko <jiri@nvidia.com>

Currently it is possible for netlink kernel user to pass custom
filter function to broadcast send function netlink_broadcast_filtered().
However, this is not exposed to multicast send and to generic
netlink users.

Extend the api and introduce a netlink helper nlmsg_multicast_filtered()
and a generic netlink helper genlmsg_multicast_netns_filtered()
to allow generic netlink families to specify filter function
while sending multicast messages.

Signed-off-by: Jiri Pirko <jiri@nvidia.com>
---
v1->v2:
- used netlink_filter_fn introduce by the previous patch
- added return comments to silence scripts/kernel-doc warnings
---
 include/net/genetlink.h | 35 +++++++++++++++++++++++++++++++----
 include/net/netlink.h   | 31 +++++++++++++++++++++++++++----
 2 files changed, 58 insertions(+), 8 deletions(-)

diff --git a/include/net/genetlink.h b/include/net/genetlink.h
index dbf11464e96a..c3fe1d5b4254 100644
--- a/include/net/genetlink.h
+++ b/include/net/genetlink.h
@@ -441,6 +441,35 @@ static inline void genlmsg_cancel(struct sk_buff *skb, void *hdr)
 		nlmsg_cancel(skb, hdr - GENL_HDRLEN - NLMSG_HDRLEN);
 }
 
+/**
+ * genlmsg_multicast_netns_filtered - multicast a netlink message
+ *				      to a specific netns with filter
+ *				      function
+ * @family: the generic netlink family
+ * @net: the net namespace
+ * @skb: netlink message as socket buffer
+ * @portid: own netlink portid to avoid sending to yourself
+ * @group: offset of multicast group in groups array
+ * @flags: allocation flags
+ * @filter: filter function
+ * @filter_data: filter function private data
+ *
+ * Return: 0 on success, negative error code for failure.
+ */
+static inline int
+genlmsg_multicast_netns_filtered(const struct genl_family *family,
+				 struct net *net, struct sk_buff *skb,
+				 u32 portid, unsigned int group, gfp_t flags,
+				 netlink_filter_fn filter,
+				 void *filter_data)
+{
+	if (WARN_ON_ONCE(group >= family->n_mcgrps))
+		return -EINVAL;
+	group = family->mcgrp_offset + group;
+	return nlmsg_multicast_filtered(net->genl_sock, skb, portid, group,
+					flags, filter, filter_data);
+}
+
 /**
  * genlmsg_multicast_netns - multicast a netlink message to a specific netns
  * @family: the generic netlink family
@@ -454,10 +483,8 @@ static inline int genlmsg_multicast_netns(const struct genl_family *family,
 					  struct net *net, struct sk_buff *skb,
 					  u32 portid, unsigned int group, gfp_t flags)
 {
-	if (WARN_ON_ONCE(group >= family->n_mcgrps))
-		return -EINVAL;
-	group = family->mcgrp_offset + group;
-	return nlmsg_multicast(net->genl_sock, skb, portid, group, flags);
+	return genlmsg_multicast_netns_filtered(family, net, skb, portid,
+						group, flags, NULL, NULL);
 }
 
 /**
diff --git a/include/net/netlink.h b/include/net/netlink.h
index 167b91348e57..2ba1438b7066 100644
--- a/include/net/netlink.h
+++ b/include/net/netlink.h
@@ -1087,27 +1087,50 @@ static inline void nlmsg_free(struct sk_buff *skb)
 }
 
 /**
- * nlmsg_multicast - multicast a netlink message
+ * nlmsg_multicast_filtered - multicast a netlink message with filter function
  * @sk: netlink socket to spread messages to
  * @skb: netlink message as socket buffer
  * @portid: own netlink portid to avoid sending to yourself
  * @group: multicast group id
  * @flags: allocation flags
+ * @filter: filter function
+ * @filter_data: filter function private data
+ *
+ * Return: 0 on success, negative error code for failure.
  */
-static inline int nlmsg_multicast(struct sock *sk, struct sk_buff *skb,
-				  u32 portid, unsigned int group, gfp_t flags)
+static inline int nlmsg_multicast_filtered(struct sock *sk, struct sk_buff *skb,
+					   u32 portid, unsigned int group,
+					   gfp_t flags,
+					   netlink_filter_fn filter,
+					   void *filter_data)
 {
 	int err;
 
 	NETLINK_CB(skb).dst_group = group;
 
-	err = netlink_broadcast(sk, skb, portid, group, flags);
+	err = netlink_broadcast_filtered(sk, skb, portid, group, flags,
+					 filter, filter_data);
 	if (err > 0)
 		err = 0;
 
 	return err;
 }
 
+/**
+ * nlmsg_multicast - multicast a netlink message
+ * @sk: netlink socket to spread messages to
+ * @skb: netlink message as socket buffer
+ * @portid: own netlink portid to avoid sending to yourself
+ * @group: multicast group id
+ * @flags: allocation flags
+ */
+static inline int nlmsg_multicast(struct sock *sk, struct sk_buff *skb,
+				  u32 portid, unsigned int group, gfp_t flags)
+{
+	return nlmsg_multicast_filtered(sk, skb, portid, group, flags,
+					NULL, NULL);
+}
+
 /**
  * nlmsg_unicast - unicast a netlink message
  * @sk: netlink socket to spread message to
-- 
2.41.0

[patch net-next v5 8/9] devlink: add a command to set notification filter and use it for multicasts

[Jiri Pirko]

From: Jiri Pirko <jiri@nvidia.com>

Currently the user listening on a socket for devlink notifications
gets always all messages for all existing instances, even if he is
interested only in one of those. That may cause unnecessary overhead
on setups with thousands of instances present.

User is currently able to narrow down the devlink objects replies
to dump commands by specifying select attributes.

Allow similar approach for notifications. Introduce a new devlink
NOTIFY_FILTER_SET which the user passes the select attributes. Store
these per-socket and use them for filtering messages
during multicast send.

Signed-off-by: Jiri Pirko <jiri@nvidia.com>
---
v4->v5:
- rebased on top of generic netlink per sock per family pointer
  allocation code
- changed the flt to be stored in family priv rcu pointer, protected by
  spin lock
- changed to use size_add() helper for kzalloc() size computation
- removed generated userspace bits
v3->v4:
- rebased on top of genl_sk_priv_*() introduction
---
 Documentation/netlink/specs/devlink.yaml |  10 +++
 include/uapi/linux/devlink.h             |   2 +
 net/devlink/devl_internal.h              |  34 ++++++-
 net/devlink/netlink.c                    | 108 +++++++++++++++++++++++
 net/devlink/netlink_gen.c                |  15 +++-
 net/devlink/netlink_gen.h                |   4 +-
 6 files changed, 169 insertions(+), 4 deletions(-)

diff --git a/Documentation/netlink/specs/devlink.yaml b/Documentation/netlink/specs/devlink.yaml
index 43067e1f63aa..6bad1d3454b7 100644
--- a/Documentation/netlink/specs/devlink.yaml
+++ b/Documentation/netlink/specs/devlink.yaml
@@ -2055,3 +2055,13 @@ operations:
             - bus-name
             - dev-name
             - selftests
+
+    -
+      name: notify-filter-set
+      doc: Set notification messages socket filter.
+      attribute-set: devlink
+      do:
+        request:
+          attributes:
+            - bus-name
+            - dev-name
diff --git a/include/uapi/linux/devlink.h b/include/uapi/linux/devlink.h
index b3c8383d342d..130cae0d3e20 100644
--- a/include/uapi/linux/devlink.h
+++ b/include/uapi/linux/devlink.h
@@ -139,6 +139,8 @@ enum devlink_command {
 	DEVLINK_CMD_SELFTESTS_GET,	/* can dump */
 	DEVLINK_CMD_SELFTESTS_RUN,
 
+	DEVLINK_CMD_NOTIFY_FILTER_SET,
+
 	/* add new commands above here */
 	__DEVLINK_CMD_MAX,
 	DEVLINK_CMD_MAX = __DEVLINK_CMD_MAX - 1
diff --git a/net/devlink/devl_internal.h b/net/devlink/devl_internal.h
index 84dc9628d3f2..82e0fb3bbebf 100644
--- a/net/devlink/devl_internal.h
+++ b/net/devlink/devl_internal.h
@@ -191,11 +191,41 @@ static inline bool devlink_nl_notify_need(struct devlink *devlink)
 				  DEVLINK_MCGRP_CONFIG);
 }
 
+struct devlink_obj_desc {
+	struct rcu_head rcu;
+	const char *bus_name;
+	const char *dev_name;
+	long data[];
+};
+
+static inline void devlink_nl_obj_desc_init(struct devlink_obj_desc *desc,
+					    struct devlink *devlink)
+{
+	memset(desc, 0, sizeof(*desc));
+	desc->bus_name = devlink->dev->bus->name;
+	desc->dev_name = dev_name(devlink->dev);
+}
+
+int devlink_nl_notify_filter(struct sock *dsk, struct sk_buff *skb, void *data);
+
+static inline void devlink_nl_notify_send_desc(struct devlink *devlink,
+					       struct sk_buff *msg,
+					       struct devlink_obj_desc *desc)
+{
+	genlmsg_multicast_netns_filtered(&devlink_nl_family,
+					 devlink_net(devlink),
+					 msg, 0, DEVLINK_MCGRP_CONFIG,
+					 GFP_KERNEL,
+					 devlink_nl_notify_filter, desc);
+}
+
 static inline void devlink_nl_notify_send(struct devlink *devlink,
 					  struct sk_buff *msg)
 {
-	genlmsg_multicast_netns(&devlink_nl_family, devlink_net(devlink),
-				msg, 0, DEVLINK_MCGRP_CONFIG, GFP_KERNEL);
+	struct devlink_obj_desc desc;
+
+	devlink_nl_obj_desc_init(&desc, devlink);
+	devlink_nl_notify_send_desc(devlink, msg, &desc);
 }
 
 /* Notify */
diff --git a/net/devlink/netlink.c b/net/devlink/netlink.c
index fa9afe3e6d9b..06f294d34a04 100644
--- a/net/devlink/netlink.c
+++ b/net/devlink/netlink.c
@@ -17,6 +17,111 @@ static const struct genl_multicast_group devlink_nl_mcgrps[] = {
 	[DEVLINK_MCGRP_CONFIG] = { .name = DEVLINK_GENL_MCGRP_CONFIG_NAME },
 };
 
+struct devlink_nl_sock_priv {
+	struct devlink_obj_desc __rcu *flt;
+	spinlock_t flt_lock; /* Protects flt. */
+};
+
+static void devlink_nl_sock_priv_init(void *priv)
+{
+	struct devlink_nl_sock_priv *sk_priv = priv;
+
+	spin_lock_init(&sk_priv->flt_lock);
+}
+
+static void devlink_nl_sock_priv_destroy(void *priv)
+{
+	struct devlink_nl_sock_priv *sk_priv = priv;
+	struct devlink_obj_desc *flt;
+
+	flt = rcu_dereference_protected(sk_priv->flt, true);
+	kfree_rcu(flt, rcu);
+}
+
+int devlink_nl_notify_filter_set_doit(struct sk_buff *skb,
+				      struct genl_info *info)
+{
+	struct devlink_nl_sock_priv *sk_priv;
+	struct nlattr **attrs = info->attrs;
+	struct devlink_obj_desc *flt;
+	size_t data_offset = 0;
+	size_t data_size = 0;
+	char *pos;
+
+	if (attrs[DEVLINK_ATTR_BUS_NAME])
+		data_size = size_add(data_size,
+				     nla_len(attrs[DEVLINK_ATTR_BUS_NAME]) + 1);
+	if (attrs[DEVLINK_ATTR_DEV_NAME])
+		data_size = size_add(data_size,
+				     nla_len(attrs[DEVLINK_ATTR_DEV_NAME]) + 1);
+
+	flt = kzalloc(size_add(sizeof(*flt), data_size), GFP_KERNEL);
+	if (!flt)
+		return -ENOMEM;
+
+	pos = (char *) flt->data;
+	if (attrs[DEVLINK_ATTR_BUS_NAME]) {
+		data_offset += nla_strscpy(pos,
+					   attrs[DEVLINK_ATTR_BUS_NAME],
+					   data_size) + 1;
+		flt->bus_name = pos;
+		pos += data_offset;
+	}
+	if (attrs[DEVLINK_ATTR_DEV_NAME]) {
+		nla_strscpy(pos, attrs[DEVLINK_ATTR_DEV_NAME],
+			    data_size - data_offset);
+		flt->dev_name = pos;
+	}
+
+	/* Don't attach empty filter. */
+	if (!flt->bus_name && !flt->dev_name) {
+		kfree(flt);
+		flt = NULL;
+	}
+
+	sk_priv = genl_sk_priv_get(NETLINK_CB(skb).sk, &devlink_nl_family);
+	if (IS_ERR(sk_priv)) {
+		kfree(flt);
+		return PTR_ERR(sk_priv);
+	}
+	spin_lock(&sk_priv->flt_lock);
+	flt = rcu_replace_pointer(sk_priv->flt, flt,
+				  lockdep_is_held(&sk_priv->flt_lock));
+	spin_unlock(&sk_priv->flt_lock);
+	kfree_rcu(flt, rcu);
+	return 0;
+}
+
+static bool devlink_obj_desc_match(const struct devlink_obj_desc *desc,
+				   const struct devlink_obj_desc *flt)
+{
+	if (desc->bus_name && flt->bus_name &&
+	    strcmp(desc->bus_name, flt->bus_name))
+		return false;
+	if (desc->dev_name && flt->dev_name &&
+	    strcmp(desc->dev_name, flt->dev_name))
+		return false;
+	return true;
+}
+
+int devlink_nl_notify_filter(struct sock *dsk, struct sk_buff *skb, void *data)
+{
+	struct devlink_obj_desc *desc = data;
+	struct devlink_nl_sock_priv *sk_priv;
+	struct devlink_obj_desc *flt;
+	int ret = 0;
+
+	rcu_read_lock();
+	sk_priv = __genl_sk_priv_get(dsk, &devlink_nl_family);
+	if (sk_priv) {
+		flt = rcu_dereference(sk_priv->flt);
+		if (flt)
+			ret = !devlink_obj_desc_match(desc, flt);
+	}
+	rcu_read_unlock();
+	return ret;
+}
+
 int devlink_nl_put_nested_handle(struct sk_buff *msg, struct net *net,
 				 struct devlink *devlink, int attrtype)
 {
@@ -256,4 +361,7 @@ struct genl_family devlink_nl_family __ro_after_init = {
 	.resv_start_op	= DEVLINK_CMD_SELFTESTS_RUN + 1,
 	.mcgrps		= devlink_nl_mcgrps,
 	.n_mcgrps	= ARRAY_SIZE(devlink_nl_mcgrps),
+	.sock_priv_size		= sizeof(struct devlink_nl_sock_priv),
+	.sock_priv_init		= devlink_nl_sock_priv_init,
+	.sock_priv_destroy	= devlink_nl_sock_priv_destroy,
 };
diff --git a/net/devlink/netlink_gen.c b/net/devlink/netlink_gen.c
index 95f9b4350ab7..1cb0e05305d2 100644
--- a/net/devlink/netlink_gen.c
+++ b/net/devlink/netlink_gen.c
@@ -560,8 +560,14 @@ static const struct nla_policy devlink_selftests_run_nl_policy[DEVLINK_ATTR_SELF
 	[DEVLINK_ATTR_SELFTESTS] = NLA_POLICY_NESTED(devlink_dl_selftest_id_nl_policy),
 };
 
+/* DEVLINK_CMD_NOTIFY_FILTER_SET - do */
+static const struct nla_policy devlink_notify_filter_set_nl_policy[DEVLINK_ATTR_DEV_NAME + 1] = {
+	[DEVLINK_ATTR_BUS_NAME] = { .type = NLA_NUL_STRING, },
+	[DEVLINK_ATTR_DEV_NAME] = { .type = NLA_NUL_STRING, },
+};
+
 /* Ops table for devlink */
-const struct genl_split_ops devlink_nl_ops[73] = {
+const struct genl_split_ops devlink_nl_ops[74] = {
 	{
 		.cmd		= DEVLINK_CMD_GET,
 		.validate	= GENL_DONT_VALIDATE_STRICT,
@@ -1233,4 +1239,11 @@ const struct genl_split_ops devlink_nl_ops[73] = {
 		.maxattr	= DEVLINK_ATTR_SELFTESTS,
 		.flags		= GENL_ADMIN_PERM | GENL_CMD_CAP_DO,
 	},
+	{
+		.cmd		= DEVLINK_CMD_NOTIFY_FILTER_SET,
+		.doit		= devlink_nl_notify_filter_set_doit,
+		.policy		= devlink_notify_filter_set_nl_policy,
+		.maxattr	= DEVLINK_ATTR_DEV_NAME,
+		.flags		= GENL_CMD_CAP_DO,
+	},
 };
diff --git a/net/devlink/netlink_gen.h b/net/devlink/netlink_gen.h
index 02f3c0bfae0e..8f2bd50ddf5e 100644
--- a/net/devlink/netlink_gen.h
+++ b/net/devlink/netlink_gen.h
@@ -16,7 +16,7 @@ extern const struct nla_policy devlink_dl_port_function_nl_policy[DEVLINK_PORT_F
 extern const struct nla_policy devlink_dl_selftest_id_nl_policy[DEVLINK_ATTR_SELFTEST_ID_FLASH + 1];
 
 /* Ops table for devlink */
-extern const struct genl_split_ops devlink_nl_ops[73];
+extern const struct genl_split_ops devlink_nl_ops[74];
 
 int devlink_nl_pre_doit(const struct genl_split_ops *ops, struct sk_buff *skb,
 			struct genl_info *info);
@@ -142,5 +142,7 @@ int devlink_nl_selftests_get_doit(struct sk_buff *skb, struct genl_info *info);
 int devlink_nl_selftests_get_dumpit(struct sk_buff *skb,
 				    struct netlink_callback *cb);
 int devlink_nl_selftests_run_doit(struct sk_buff *skb, struct genl_info *info);
+int devlink_nl_notify_filter_set_doit(struct sk_buff *skb,
+				      struct genl_info *info);
 
 #endif /* _LINUX_DEVLINK_GEN_H */
-- 
2.41.0

[patch net-next v5 9/9] devlink: extend multicast filtering by port index

[Jiri Pirko]

From: Jiri Pirko <jiri@nvidia.com>

Expose the previously introduced notification multicast messages
filtering infrastructure and allow the user to select messages using
port index.

Signed-off-by: Jiri Pirko <jiri@nvidia.com>
---
v4->v5:
- removed generated userspace bits
v3->v4:
- rebased on top of genl_sk_priv_*() introduction
---
 Documentation/netlink/specs/devlink.yaml |  1 +
 net/devlink/devl_internal.h              |  9 +++++++++
 net/devlink/health.c                     |  6 +++++-
 net/devlink/netlink.c                    | 10 +++++++++-
 net/devlink/netlink_gen.c                |  5 +++--
 net/devlink/port.c                       |  5 ++++-
 6 files changed, 31 insertions(+), 5 deletions(-)

diff --git a/Documentation/netlink/specs/devlink.yaml b/Documentation/netlink/specs/devlink.yaml
index 6bad1d3454b7..4996ff7e09b6 100644
--- a/Documentation/netlink/specs/devlink.yaml
+++ b/Documentation/netlink/specs/devlink.yaml
@@ -2065,3 +2065,4 @@ operations:
           attributes:
             - bus-name
             - dev-name
+            - port-index
diff --git a/net/devlink/devl_internal.h b/net/devlink/devl_internal.h
index 82e0fb3bbebf..c7a8e13f917c 100644
--- a/net/devlink/devl_internal.h
+++ b/net/devlink/devl_internal.h
@@ -195,6 +195,8 @@ struct devlink_obj_desc {
 	struct rcu_head rcu;
 	const char *bus_name;
 	const char *dev_name;
+	unsigned int port_index;
+	bool port_index_valid;
 	long data[];
 };
 
@@ -206,6 +208,13 @@ static inline void devlink_nl_obj_desc_init(struct devlink_obj_desc *desc,
 	desc->dev_name = dev_name(devlink->dev);
 }
 
+static inline void devlink_nl_obj_desc_port_set(struct devlink_obj_desc *desc,
+						struct devlink_port *devlink_port)
+{
+	desc->port_index = devlink_port->index;
+	desc->port_index_valid = true;
+}
+
 int devlink_nl_notify_filter(struct sock *dsk, struct sk_buff *skb, void *data);
 
 static inline void devlink_nl_notify_send_desc(struct devlink *devlink,
diff --git a/net/devlink/health.c b/net/devlink/health.c
index 1d59ec0202f6..acb8c0e174bb 100644
--- a/net/devlink/health.c
+++ b/net/devlink/health.c
@@ -490,6 +490,7 @@ static void devlink_recover_notify(struct devlink_health_reporter *reporter,
 				   enum devlink_command cmd)
 {
 	struct devlink *devlink = reporter->devlink;
+	struct devlink_obj_desc desc;
 	struct sk_buff *msg;
 	int err;
 
@@ -509,7 +510,10 @@ static void devlink_recover_notify(struct devlink_health_reporter *reporter,
 		return;
 	}
 
-	devlink_nl_notify_send(devlink, msg);
+	devlink_nl_obj_desc_init(&desc, devlink);
+	if (reporter->devlink_port)
+		devlink_nl_obj_desc_port_set(&desc, reporter->devlink_port);
+	devlink_nl_notify_send_desc(devlink, msg, &desc);
 }
 
 void
diff --git a/net/devlink/netlink.c b/net/devlink/netlink.c
index 06f294d34a04..c6de311130fb 100644
--- a/net/devlink/netlink.c
+++ b/net/devlink/netlink.c
@@ -73,8 +73,13 @@ int devlink_nl_notify_filter_set_doit(struct sk_buff *skb,
 		flt->dev_name = pos;
 	}
 
+	if (attrs[DEVLINK_ATTR_PORT_INDEX]) {
+		flt->port_index = nla_get_u32(attrs[DEVLINK_ATTR_PORT_INDEX]);
+		flt->port_index_valid = true;
+	}
+
 	/* Don't attach empty filter. */
-	if (!flt->bus_name && !flt->dev_name) {
+	if (!flt->bus_name && !flt->dev_name && !flt->port_index_valid) {
 		kfree(flt);
 		flt = NULL;
 	}
@@ -101,6 +106,9 @@ static bool devlink_obj_desc_match(const struct devlink_obj_desc *desc,
 	if (desc->dev_name && flt->dev_name &&
 	    strcmp(desc->dev_name, flt->dev_name))
 		return false;
+	if (desc->port_index_valid && flt->port_index_valid &&
+	    desc->port_index != flt->port_index)
+		return false;
 	return true;
 }
 
diff --git a/net/devlink/netlink_gen.c b/net/devlink/netlink_gen.c
index 1cb0e05305d2..c81cf2dd154f 100644
--- a/net/devlink/netlink_gen.c
+++ b/net/devlink/netlink_gen.c
@@ -561,9 +561,10 @@ static const struct nla_policy devlink_selftests_run_nl_policy[DEVLINK_ATTR_SELF
 };
 
 /* DEVLINK_CMD_NOTIFY_FILTER_SET - do */
-static const struct nla_policy devlink_notify_filter_set_nl_policy[DEVLINK_ATTR_DEV_NAME + 1] = {
+static const struct nla_policy devlink_notify_filter_set_nl_policy[DEVLINK_ATTR_PORT_INDEX + 1] = {
 	[DEVLINK_ATTR_BUS_NAME] = { .type = NLA_NUL_STRING, },
 	[DEVLINK_ATTR_DEV_NAME] = { .type = NLA_NUL_STRING, },
+	[DEVLINK_ATTR_PORT_INDEX] = { .type = NLA_U32, },
 };
 
 /* Ops table for devlink */
@@ -1243,7 +1244,7 @@ const struct genl_split_ops devlink_nl_ops[74] = {
 		.cmd		= DEVLINK_CMD_NOTIFY_FILTER_SET,
 		.doit		= devlink_nl_notify_filter_set_doit,
 		.policy		= devlink_notify_filter_set_nl_policy,
-		.maxattr	= DEVLINK_ATTR_DEV_NAME,
+		.maxattr	= DEVLINK_ATTR_PORT_INDEX,
 		.flags		= GENL_CMD_CAP_DO,
 	},
 };
diff --git a/net/devlink/port.c b/net/devlink/port.c
index 758df3000a1b..62e54e152ecf 100644
--- a/net/devlink/port.c
+++ b/net/devlink/port.c
@@ -507,6 +507,7 @@ static void devlink_port_notify(struct devlink_port *devlink_port,
 				enum devlink_command cmd)
 {
 	struct devlink *devlink = devlink_port->devlink;
+	struct devlink_obj_desc desc;
 	struct sk_buff *msg;
 	int err;
 
@@ -525,7 +526,9 @@ static void devlink_port_notify(struct devlink_port *devlink_port,
 		return;
 	}
 
-	devlink_nl_notify_send(devlink, msg);
+	devlink_nl_obj_desc_init(&desc, devlink);
+	devlink_nl_obj_desc_port_set(&desc, devlink_port);
+	devlink_nl_notify_send_desc(devlink, msg, &desc);
 }
 
 static void devlink_ports_notify(struct devlink *devlink,
-- 
2.41.0

Re: [patch net-next v5 5/9] genetlink: introduce per-sock family private storage

[Jakub Kicinski]

On Wed,  6 Dec 2023 19:21:16 +0100 Jiri Pirko wrote:
> diff --git a/include/net/genetlink.h b/include/net/genetlink.h
> index e18a4c0d69ee..dbf11464e96a 100644
> --- a/include/net/genetlink.h
> +++ b/include/net/genetlink.h
> @@ -87,6 +87,9 @@ struct genl_family {
>  	int			id;
>  	/* starting number of multicast group IDs in this family */
>  	unsigned int		mcgrp_offset;
> +	size_t			sock_priv_size;
> +	void			(*sock_priv_init)(void *priv);
> +	void			(*sock_priv_destroy)(void *priv);

👍️

but I think it should be above the private fields (and have kdoc)
The families are expected to make use the new fields, and are not
supposed to touch anything private.

> --- a/net/netlink/af_netlink.h
> +++ b/net/netlink/af_netlink.h
> @@ -60,6 +60,21 @@ static inline struct netlink_sock *nlk_sk(struct sock *sk)
>  
>  #define nlk_test_bit(nr, sk) test_bit(NETLINK_F_##nr, &nlk_sk(sk)->flags)
>  
> +struct genl_sock {
> +	struct netlink_sock nlk_sk;
> +	struct xarray *family_privs;
> +};
> +
> +static inline struct genl_sock *genl_sk(struct sock *sk)
> +{
> +	return container_of(nlk_sk(sk), struct genl_sock, nlk_sk);
> +}
> +
> +/* Size of netlink sock is size of the biggest user with priv,
> + * which is currently just Generic Netlink.
> + */
> +#define NETLINK_SOCK_SIZE sizeof(struct genl_sock)

Would feel a little cleaner to me to add

#define NETLINK_SOCK_PROTO_SIZE		8

add that to the size, build time check that struct genl_sock's
size is <= than sizeof(struct netlink_sock) + NETLINK_SOCK_PROTO_SIZE

This way we don't have to fumble the layering by putting genl stuff
in af_netlink.h

> +struct genl_sk_priv {
> +	void (*destructor)(void *priv);
> +	long priv[];
> +};
> +
> +static struct genl_sk_priv *genl_sk_priv_alloc(struct genl_family *family)
> +{
> +	struct genl_sk_priv *priv;
> +
> +	priv = kzalloc(size_add(sizeof(*priv), family->sock_priv_size),
> +		       GFP_KERNEL);
> +	if (!priv)
> +		return ERR_PTR(-ENOMEM);
> +	priv->destructor = family->sock_priv_destroy;

family->sock_priv_destroy may be in module memory.
I think you need to wipe them when family goes :(

> +	if (family->sock_priv_init)
> +		family->sock_priv_init(priv->priv);
> +	return priv;
> +}

> +static struct xarray *genl_family_privs_get(struct genl_sock *gsk)
> +{
> +	struct xarray *family_privs;
> +
> +again:
> +	family_privs = READ_ONCE(gsk->family_privs);
> +	if (family_privs)
> +		return family_privs;
> +
> +	family_privs = kzalloc(sizeof(*family_privs), GFP_KERNEL);
> +	if (!family_privs)
> +		return ERR_PTR(-ENOMEM);
> +	xa_init_flags(family_privs, XA_FLAGS_ALLOC);
> +
> +	/* Use genl lock to protect family_privs to be
> +	 * initialized in parallel by different CPU.
> +	 */
> +	genl_lock();
> +	if (unlikely(gsk->family_privs)) {
> +		xa_destroy(family_privs);
> +		kfree(family_privs);
> +		genl_unlock();

nit: unlock can be moved up

> +		goto again;

why not return READ_ONCE(gsk->family_privs); ?
there's no need to loop

One could also be tempted to:

lock()
if (likely(!gsk->family_privs)) {
	WRITE
} else {
	destory()
	free()
	family_privs = READ
}
unlock()

but it could be argued success path should be flat

> +	}
> +	WRITE_ONCE(gsk->family_privs, family_privs);
> +	genl_unlock();
> +	return family_privs;
> +}

Re: [patch net-next v5 5/9] genetlink: introduce per-sock family private storage

[Jiri Pirko]

Fri, Dec 08, 2023 at 03:55:26AM CET, kuba@kernel.org wrote:
>On Wed,  6 Dec 2023 19:21:16 +0100 Jiri Pirko wrote:
>> diff --git a/include/net/genetlink.h b/include/net/genetlink.h
>> index e18a4c0d69ee..dbf11464e96a 100644
>> --- a/include/net/genetlink.h
>> +++ b/include/net/genetlink.h
>> @@ -87,6 +87,9 @@ struct genl_family {
>>  	int			id;
>>  	/* starting number of multicast group IDs in this family */
>>  	unsigned int		mcgrp_offset;
>> +	size_t			sock_priv_size;
>> +	void			(*sock_priv_init)(void *priv);
>> +	void			(*sock_priv_destroy)(void *priv);
>
>👍️
>
>but I think it should be above the private fields (and have kdoc)
>The families are expected to make use the new fields, and are not
>supposed to touch anything private.

Oh, right, good point, I missed that.


>
>> --- a/net/netlink/af_netlink.h
>> +++ b/net/netlink/af_netlink.h
>> @@ -60,6 +60,21 @@ static inline struct netlink_sock *nlk_sk(struct sock *sk)
>>  
>>  #define nlk_test_bit(nr, sk) test_bit(NETLINK_F_##nr, &nlk_sk(sk)->flags)
>>  
>> +struct genl_sock {
>> +	struct netlink_sock nlk_sk;
>> +	struct xarray *family_privs;
>> +};
>> +
>> +static inline struct genl_sock *genl_sk(struct sock *sk)
>> +{
>> +	return container_of(nlk_sk(sk), struct genl_sock, nlk_sk);
>> +}
>> +
>> +/* Size of netlink sock is size of the biggest user with priv,
>> + * which is currently just Generic Netlink.
>> + */
>> +#define NETLINK_SOCK_SIZE sizeof(struct genl_sock)
>
>Would feel a little cleaner to me to add
>
>#define NETLINK_SOCK_PROTO_SIZE		8
>
>add that to the size, build time check that struct genl_sock's
>size is <= than sizeof(struct netlink_sock) + NETLINK_SOCK_PROTO_SIZE
>
>This way we don't have to fumble the layering by putting genl stuff
>in af_netlink.h

Yeah, I had it like that originally, I didn't like it :) Mainly because
if someone adds-in another field in the future, the build time check
may only fail on some archs. Also, wasting memory on archs there pointer
is 4 bytes :) But as you wish, I don't mind to switch it back.


>
>> +struct genl_sk_priv {
>> +	void (*destructor)(void *priv);
>> +	long priv[];
>> +};
>> +
>> +static struct genl_sk_priv *genl_sk_priv_alloc(struct genl_family *family)
>> +{
>> +	struct genl_sk_priv *priv;
>> +
>> +	priv = kzalloc(size_add(sizeof(*priv), family->sock_priv_size),
>> +		       GFP_KERNEL);
>> +	if (!priv)
>> +		return ERR_PTR(-ENOMEM);
>> +	priv->destructor = family->sock_priv_destroy;
>
>family->sock_priv_destroy may be in module memory.
>I think you need to wipe them when family goes :(
>
>> +	if (family->sock_priv_init)
>> +		family->sock_priv_init(priv->priv);
>> +	return priv;
>> +}
>
>> +static struct xarray *genl_family_privs_get(struct genl_sock *gsk)
>> +{
>> +	struct xarray *family_privs;
>> +
>> +again:
>> +	family_privs = READ_ONCE(gsk->family_privs);
>> +	if (family_privs)
>> +		return family_privs;
>> +
>> +	family_privs = kzalloc(sizeof(*family_privs), GFP_KERNEL);
>> +	if (!family_privs)
>> +		return ERR_PTR(-ENOMEM);
>> +	xa_init_flags(family_privs, XA_FLAGS_ALLOC);
>> +
>> +	/* Use genl lock to protect family_privs to be
>> +	 * initialized in parallel by different CPU.
>> +	 */
>> +	genl_lock();
>> +	if (unlikely(gsk->family_privs)) {
>> +		xa_destroy(family_privs);
>> +		kfree(family_privs);
>> +		genl_unlock();
>
>nit: unlock can be moved up

Okay.


>
>> +		goto again;
>
>why not return READ_ONCE(gsk->family_privs); ?
>there's no need to loop

Right.

>
>One could also be tempted to:
>
>lock()
>if (likely(!gsk->family_privs)) {
>	WRITE
>} else {
>	destory()
>	free()
>	family_privs = READ
>}
>unlock()
>
>but it could be argued success path should be flat

Okay, I will think about it.



Thanks!


>
>> +	}
>> +	WRITE_ONCE(gsk->family_privs, family_privs);
>> +	genl_unlock();
>> +	return family_privs;
>> +}

Re: [patch net-next v5 5/9] genetlink: introduce per-sock family private storage

[Jiri Pirko]

Fri, Dec 08, 2023 at 03:55:26AM CET, kuba@kernel.org wrote:
>On Wed,  6 Dec 2023 19:21:16 +0100 Jiri Pirko wrote:

[...]

>> +static struct genl_sk_priv *genl_sk_priv_alloc(struct genl_family *family)
>> +{
>> +	struct genl_sk_priv *priv;
>> +
>> +	priv = kzalloc(size_add(sizeof(*priv), family->sock_priv_size),
>> +		       GFP_KERNEL);
>> +	if (!priv)
>> +		return ERR_PTR(-ENOMEM);
>> +	priv->destructor = family->sock_priv_destroy;
>
>family->sock_priv_destroy may be in module memory.
>I think you need to wipe them when family goes :(

Crap. That's a bit problematic. Family can unregister and register
again, with user having the same sock sill opened with legitimate
expectation of filter being applied. Don't see now how to handle this
other then no-destroy and just kfree here in genetlink.c :/ Going back
to v4?


>
>> +	if (family->sock_priv_init)
>> +		family->sock_priv_init(priv->priv);
>> +	return priv;
>> +}

[...]

Re: [patch net-next v5 5/9] genetlink: introduce per-sock family private storage

[Jakub Kicinski]

On Fri, 8 Dec 2023 15:21:52 +0100 Jiri Pirko wrote:
> >> +static struct genl_sk_priv *genl_sk_priv_alloc(struct genl_family *family)
> >> +{
> >> +	struct genl_sk_priv *priv;
> >> +
> >> +	priv = kzalloc(size_add(sizeof(*priv), family->sock_priv_size),
> >> +		       GFP_KERNEL);
> >> +	if (!priv)
> >> +		return ERR_PTR(-ENOMEM);
> >> +	priv->destructor = family->sock_priv_destroy;  
> >
> >family->sock_priv_destroy may be in module memory.
> >I think you need to wipe them when family goes :(  
> 
> Crap. That's a bit problematic. Family can unregister and register
> again, with user having the same sock sill opened with legitimate
> expectation of filter being applied. Don't see now how to handle this
> other then no-destroy and just kfree here in genetlink.c :/ Going back
> to v4?

When family gets removed all subs must be cleared. So the user
sock will have to resolve the mcast ID again, and re-subscribe
again to get any notification. Having to re-sub implies having
to re-add filters in my mind.

Re: [patch net-next v5 5/9] genetlink: introduce per-sock family private storage

[Jiri Pirko]

Fri, Dec 08, 2023 at 05:11:23PM CET, kuba@kernel.org wrote:
>On Fri, 8 Dec 2023 15:21:52 +0100 Jiri Pirko wrote:
>> >> +static struct genl_sk_priv *genl_sk_priv_alloc(struct genl_family *family)
>> >> +{
>> >> +	struct genl_sk_priv *priv;
>> >> +
>> >> +	priv = kzalloc(size_add(sizeof(*priv), family->sock_priv_size),
>> >> +		       GFP_KERNEL);
>> >> +	if (!priv)
>> >> +		return ERR_PTR(-ENOMEM);
>> >> +	priv->destructor = family->sock_priv_destroy;  
>> >
>> >family->sock_priv_destroy may be in module memory.
>> >I think you need to wipe them when family goes :(  
>> 
>> Crap. That's a bit problematic. Family can unregister and register
>> again, with user having the same sock sill opened with legitimate
>> expectation of filter being applied. Don't see now how to handle this
>> other then no-destroy and just kfree here in genetlink.c :/ Going back
>> to v4?
>
>When family gets removed all subs must be cleared. So the user
>sock will have to resolve the mcast ID again, and re-subscribe
>again to get any notification. Having to re-sub implies having
>to re-add filters in my mind.

Okay, that sounds fine. Thanks!