Skip to contentSkip to conversation

rtnetlink: make the "split" NLM_DONE handling generic

PATCH Applied Jakub Kicinski 4 participants 5 messages Jun 3, 2024 – Jun 5, 2024

Subsystems (3)
lore
Share & tools
lore

Jakub Kicinski

author maintainer
#0 lore raw

PATCH v3 1/1 rtnetlink: make the "split" NLM_DONE handling generic

Jaroslav reports Dell's OMSA Systems Management Data Engine expects NLM_DONE in a separate recvmsg(), both for rtnl_dump_ifinfo() and inet_dump_ifaddr(). We already added a similar fix previously in commit 460b0d33cf10 ("inet: bring NLM_DONE out to a separate recv() again")

Instead of modifying all the dump handlers, and making them look different than modern for_each_netdev_dump()-based dump handlers - put the workaround in rtnetlink code. This will also help us move the custom rtnl-locking from af_netlink in the future (in net-next).

Note that this change is not touching rtnl_dump_all(). rtnl_dump_all() is different kettle of fish and a potential problem. We now mix families in a single recvmsg(), but NLM_DONE is not coalesced.

Tested:

  ./cli.py --dbg-small-recv 4096 --spec netlink/specs/rt_addr.yaml \
           --dump getaddr --json '{"ifa-family": 2}'

  ./cli.py --dbg-small-recv 4096 --spec netlink/specs/rt_route.yaml \
           --dump getroute --json '{"rtm-family": 2}'

  ./cli.py --dbg-small-recv 4096 --spec netlink/specs/rt_link.yaml \
           --dump getlink
Fixes 3e41af90767d ("rtnetlink: use xarray iterator to implement rtnl_dump_ifinfo()")
Fixes cdb2f80f1c10 ("inet: use xa_array iterator to implement inet_dump_ifaddr()")

---
v3:
- change the approach to centralized handling + flag
v2:
- adjust the comment in the kdoc
- add getlink

---
 include/net/rtnetlink.h |  1 +
 net/core/rtnetlink.c    | 44 +++++++++++++++++++++++++++++++++++++++--
 net/ipv4/devinet.c      |  2 +-
 net/ipv4/fib_frontend.c |  7 +------
 4 files changed, 45 insertions(+), 9 deletions(-)
include/net/rtnetlink.h +1 -0
index 3bfb80bad173..b45d57b5968a 100644
@@ -13,6 +13,7 @@ enum rtnl_link_flags {
 	RTNL_FLAG_DOIT_UNLOCKED		= BIT(0),
 	RTNL_FLAG_BULK_DEL_SUPPORTED	= BIT(1),
 	RTNL_FLAG_DUMP_UNLOCKED		= BIT(2),
+	RTNL_FLAG_DUMP_SPLIT_NLM_DONE	= BIT(3),	/* legacy behavior */
 };
 
 enum rtnl_kinds {
net/core/rtnetlink.c +42 -2

62 lines not on this page · raw message

net/ipv4/devinet.c +1 -1
index f3892ee9dfb3..d09f557eaa77 100644
@@ -2805,7 +2805,7 @@ void __init devinet_init(void)
 	rtnl_register(PF_INET, RTM_NEWADDR, inet_rtm_newaddr, NULL, 0);
 	rtnl_register(PF_INET, RTM_DELADDR, inet_rtm_deladdr, NULL, 0);
 	rtnl_register(PF_INET, RTM_GETADDR, NULL, inet_dump_ifaddr,
-		      RTNL_FLAG_DUMP_UNLOCKED);
+		      RTNL_FLAG_DUMP_UNLOCKED | RTNL_FLAG_DUMP_SPLIT_NLM_DONE);
 	rtnl_register(PF_INET, RTM_GETNETCONF, inet_netconf_get_devconf,
 		      inet_netconf_dump_devconf,
 		      RTNL_FLAG_DOIT_UNLOCKED | RTNL_FLAG_DUMP_UNLOCKED);
net/ipv4/fib_frontend.c +1 -6
index c484b1c0fc00..7ad2cafb9276 100644
@@ -1050,11 +1050,6 @@ static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 			e++;
 		}
 	}
-
-	/* Don't let NLM_DONE coalesce into a message, even if it could.
-	 * Some user space expects NLM_DONE in a separate recv().
-	 */
-	err = skb->len;
 out:
 
 	cb->args[1] = e;
@@ -1665,5 +1660,5 @@ void __init ip_fib_init(void)
 	rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, 0);
 	rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, 0);
 	rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib,
-		      RTNL_FLAG_DUMP_UNLOCKED);
+		      RTNL_FLAG_DUMP_UNLOCKED | RTNL_FLAG_DUMP_SPLIT_NLM_DONE);
 }
Signature · 1 line
2.45.1
Reviewed-by on v3 1/1
#1 lore raw
Quoted 26 lines · Jakub Kicinski

Jaroslav reports Dell's OMSA Systems Management Data Engine expects NLM_DONE in a separate recvmsg(), both for rtnl_dump_ifinfo() and inet_dump_ifaddr(). We already added a similar fix previously in commit 460b0d33cf10 ("inet: bring NLM_DONE out to a separate recv() again")

Instead of modifying all the dump handlers, and making them look different than modern for_each_netdev_dump()-based dump handlers - put the workaround in rtnetlink code. This will also help us move the custom rtnl-locking from af_netlink in the future (in net-next).

Note that this change is not touching rtnl_dump_all(). rtnl_dump_all() is different kettle of fish and a potential problem. We now mix families in a single recvmsg(), but NLM_DONE is not coalesced.

Tested:

  ./cli.py --dbg-small-recv 4096 --spec netlink/specs/rt_addr.yaml \
           --dump getaddr --json '{"ifa-family": 2}'

  ./cli.py --dbg-small-recv 4096 --spec netlink/specs/rt_route.yaml \
           --dump getroute --json '{"rtm-family": 2}'

  ./cli.py --dbg-small-recv 4096 --spec netlink/specs/rt_link.yaml \
           --dump getlink
Quoted 5 lines
Fixes 3e41af90767d ("rtnetlink: use xarray iterator to implement rtnl_dump_ifinfo()")
Fixes cdb2f80f1c10 ("inet: use xa_array iterator to implement inet_dump_ifaddr()")

Very nice, thanks a lot for taking care of this Jakub.

maintainer
#2 lore raw
In reply to Jakub Kicinski
Quoted 39 lines · Jakub Kicinski
net/core/rtnetlink.c +1 -1
@@ -6694,7 +6734,7 @@ void __init rtnetlink_init(void)
 	register_netdevice_notifier(&rtnetlink_dev_notifier);
 
 	rtnl_register(PF_UNSPEC, RTM_GETLINK, rtnl_getlink,
-		      rtnl_dump_ifinfo, 0);
+		      rtnl_dump_ifinfo, RTNL_FLAG_DUMP_SPLIT_NLM_DONE);
 	rtnl_register(PF_UNSPEC, RTM_SETLINK, rtnl_setlink, NULL, 0);
 	rtnl_register(PF_UNSPEC, RTM_NEWLINK, rtnl_newlink, NULL, 0);
 	rtnl_register(PF_UNSPEC, RTM_DELLINK, rtnl_dellink, NULL, 0);
net/ipv4/devinet.c +1 -1
index f3892ee9dfb3..d09f557eaa77 100644
@@ -2805,7 +2805,7 @@ void __init devinet_init(void)
 	rtnl_register(PF_INET, RTM_NEWADDR, inet_rtm_newaddr, NULL, 0);
 	rtnl_register(PF_INET, RTM_DELADDR, inet_rtm_deladdr, NULL, 0);
 	rtnl_register(PF_INET, RTM_GETADDR, NULL, inet_dump_ifaddr,
-		      RTNL_FLAG_DUMP_UNLOCKED);
+		      RTNL_FLAG_DUMP_UNLOCKED | RTNL_FLAG_DUMP_SPLIT_NLM_DONE);
 	rtnl_register(PF_INET, RTM_GETNETCONF, inet_netconf_get_devconf,
 		      inet_netconf_dump_devconf,
 		      RTNL_FLAG_DOIT_UNLOCKED | RTNL_FLAG_DUMP_UNLOCKED);
net/ipv4/fib_frontend.c +0 -5
index c484b1c0fc00..7ad2cafb9276 100644
@@ -1050,11 +1050,6 @@ static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 			e++;
 		}
 	}
-
-	/* Don't let NLM_DONE coalesce into a message, even if it could.
-	 * Some user space expects NLM_DONE in a separate recv().
-	 */
-	err = skb->len;
 out:
 
 	cb->args[1] = e;

net/ipv4/fib_frontend.c void __init ip_fib_init(void) · line 1664

Quoted 6 lines
net/ipv4/fib_frontend.c +1 -1
@@ -1665,5 +1660,5 @@ void __init ip_fib_init(void)
 	rtnl_register(PF_INET, RTM_NEWROUTE, inet_rtm_newroute, NULL, 0);
 	rtnl_register(PF_INET, RTM_DELROUTE, inet_rtm_delroute, NULL, 0);
 	rtnl_register(PF_INET, RTM_GETROUTE, NULL, inet_dump_fib,
-		      RTNL_FLAG_DUMP_UNLOCKED);
+		      RTNL_FLAG_DUMP_UNLOCKED | RTNL_FLAG_DUMP_SPLIT_NLM_DONE);
 }

You are using the legacy way for route, link and address dumps with the option to quickly add the others as needed (meaning when a regression is reported). If those 3 need the workaround, I think there are high odds all of the other rtnl_register users will need it. So, if there is not going to be a per-app opt-in to a new way, you might as well make this the default for all rtnl families - sadly.

author maintainer
#3 lore raw
Quoted 6 lines · On Mon, 3 Jun 2024 20:33:50 -0600 David Ahern wrote:

You are using the legacy way for route, link and address dumps with the option to quickly add the others as needed (meaning when a regression is reported). If those 3 need the workaround, I think there are high odds all of the other rtnl_register users will need it. So, if there is not going to be a per-app opt-in to a new way, you might as well make this the default for all rtnl families - sadly.

I prefer the default (for new code) to be "modern".

#4 lore raw
In reply to Jakub Kicinski
This patch was applied to netdev/net.git (main) by David S. Miller <[email protected]>:

Hello:

This patch was applied to netdev/net.git (main)
by David S. Miller <[email protected]>:

Quoted 11 lines · On Mon, 3 Jun 2024 11:48:26 -0700 you wrote:

Jaroslav reports Dell's OMSA Systems Management Data Engine expects NLM_DONE in a separate recvmsg(), both for rtnl_dump_ifinfo() and inet_dump_ifaddr(). We already added a similar fix previously in commit 460b0d33cf10 ("inet: bring NLM_DONE out to a separate recv() again")

Instead of modifying all the dump handlers, and making them look different than modern for_each_netdev_dump()-based dump handlers - put the workaround in rtnetlink code. This will also help us move the custom rtnl-locking from af_netlink in the future (in net-next).

[...]

Here is the summary with links:
- [net,v3] rtnetlink: make the "split" NLM_DONE handling generic

    https://git.kernel.org/netdev/net/c/5b4b62a169e1

You are awesome, thank you!

Signature · 2 lines
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html