When there's a large number of interfaces (e.g. vlans), teamd loses link notifications as it cannot read them as fast as kernel is broadcasting them. This often prevents teamd starting properly if started concurrently when other links are being set up. It can also fail when it's up and running, especially in the cases where the team device itself has a lot of vlans under it.
This can easily be reproduces by simple example (in SMP system) by manually adding team device with a bunch of vlans, putting it up, and starting teamd with --take-over option:
root@debian:~# ip link add name team0 type team root@debian:~# for i in `seq 100 150` ; do
ip link add link team0 name team0.$i type vlan id $i ; done
root@debian:~# ip link set team0 up root@debian:~# cat teamd.conf { "device": "team0", "runner": { "name": "activebackup" }, "ports": { "eth1": {}, "eth2": {} } } root@debian:~# teamd -o -N -f teamd.conf
At this point, teamd will not give any error messages or other indication that something is wrong. But state will not look healthy:
root@debian:~# teamdctl team0 state setup: runner: activebackup ports: eth1 link watches: link summary: up instance[link_watch_0]: name: ethtool link: up down count: 0 Failed to parse JSON port dump. command call failed (Invalid argument)
If checking state dump, it will show that port eth2 is missing info. Running strace to teamd will reveal that there's one recvmsgs() that returned -1 with errno ENOBUFS. What happened in this example was that when teamd started, all vlans got carrier up, and kernel flooded notifications faster than teamd could read them. It then lost events related to port eth2 getting enslaved and up.
The socket that joins RTNLGRP_LINK notifications uses default libnl 32k buffer size. Netlink messages are large (over 1k), and this buffer gets easily full. Kernel neither knows nor cares were notification broadcasts delivered. This cannot be fixed by simply increasing the buffer size, as there's no size that is guaranteed to work in every use case, and this can require several megabytes of buffer (a way over normal rmem_max limit) if there are hunderds of vlans.
Only way to recover from this is to refresh all ifinfo list, as it's invalidated at this point. It cannot easily work around of this by just refreshing team device and its ports, because library side might not have ports linked due to events missed, and it doesn't know about teamd configuration.
Checks now return value of nl_recvmsgs_default() for event socket. In case of ENOBUFS (which libnl nicely changes to ENOMEM), refreshes all ifinfo list. get_ifinfo_list() also checks now for removed interfaces in case of missed dellink event. Currently all TEAM_IFINFO_CHANGE handlers processed events one by one, so it had to be changed to support multiple ifinfo changes. For this, ifinfo changed flags are cleared and removed entries destroyed only after all handlers have been called.
Also, increased nl_cli.sock_event receive buffers to 96k like all other sockets. Added possibility to change this via environment variable.
Signed-off-by: Antti Tiainen atiainen@forcepoint.com --- libteam/ifinfo.c | 33 ++++++++++++++++++++++-------- libteam/libteam.c | 52 +++++++++++++++++++++++++++++++++++++++++++++--- libteam/team_private.h | 3 +++ 3 files changed, 77 insertions(+), 11 deletions(-)
diff --git a/libteam/ifinfo.c b/libteam/ifinfo.c index 72155ae..5c32a9c 100644 --- a/libteam/ifinfo.c +++ b/libteam/ifinfo.c @@ -72,6 +72,10 @@ struct team_ifinfo { #define CHANGED_PHYS_PORT_ID (1 << 5) #define CHANGED_PHYS_PORT_ID_LEN (1 << 6) #define CHANGED_ADMIN_STATE (1 << 7) +/* This is only used when tagging interfaces for finding + * removed, and thus not included to CHANGED_ANY. + */ +#define CHANGED_REFRESHING (1 << 8) #define CHANGED_ANY (CHANGED_REMOVED | CHANGED_HWADDR | \ CHANGED_HWADDR_LEN | CHANGED_IFNAME | \ CHANGED_MASTER_IFINDEX | CHANGED_PHYS_PORT_ID | \ @@ -202,7 +206,7 @@ static struct team_ifinfo *ifinfo_find(struct team_handle *th, uint32_t ifindex) return NULL; }
-static void clear_last_changed(struct team_handle *th) +void ifinfo_clear_changed(struct team_handle *th) { struct team_ifinfo *ifinfo;
@@ -236,7 +240,7 @@ static void ifinfo_destroy(struct team_ifinfo *ifinfo) free(ifinfo); }
-static void ifinfo_destroy_removed(struct team_handle *th) +void ifinfo_destroy_removed(struct team_handle *th) { struct team_ifinfo *ifinfo, *tmp;
@@ -254,8 +258,6 @@ static void obj_input_newlink(struct nl_object *obj, void *arg, bool event) uint32_t ifindex; int err;
- ifinfo_destroy_removed(th); - link = (struct rtnl_link *) obj;
ifindex = rtnl_link_get_ifindex(link); @@ -269,7 +271,7 @@ static void obj_input_newlink(struct nl_object *obj, void *arg, bool event) return; }
- clear_last_changed(th); + clear_changed(ifinfo); ifinfo_update(ifinfo, link);
if (event) @@ -292,8 +294,6 @@ static void event_handler_obj_input_dellink(struct nl_object *obj, void *arg) uint32_t ifindex; int err;
- ifinfo_destroy_removed(th); - link = (struct rtnl_link *) obj;
ifindex = rtnl_link_get_ifindex(link); @@ -311,7 +311,7 @@ static void event_handler_obj_input_dellink(struct nl_object *obj, void *arg) return; }
- clear_last_changed(th); + clear_changed(ifinfo); set_changed(ifinfo, CHANGED_REMOVED); set_call_change_handlers(th, TEAM_IFINFO_CHANGE); } @@ -367,6 +367,14 @@ int get_ifinfo_list(struct team_handle *th) }; int ret; int retry = 1; + struct team_ifinfo *ifinfo; + + /* Tag all ifinfo, this is cleared in newlink handler. + * Any interface that has this after dump is processed + * has been removed. + */ + list_for_each_node_entry(ifinfo, &th->ifinfo_list, list) + set_changed(ifinfo, CHANGED_REFRESHING);
while (retry) { retry = 0; @@ -395,6 +403,15 @@ int get_ifinfo_list(struct team_handle *th) retry = 1; } } + + list_for_each_node_entry(ifinfo, &th->ifinfo_list, list) { + if (is_changed(ifinfo, CHANGED_REFRESHING)) { + clear_changed(ifinfo); + set_changed(ifinfo, CHANGED_REMOVED); + set_call_change_handlers(th, TEAM_IFINFO_CHANGE); + } + } + ret = check_call_change_handlers(th, TEAM_IFINFO_CHANGE); if (ret < 0) err(th, "get_ifinfo_list: check_call_change_handers failed"); diff --git a/libteam/libteam.c b/libteam/libteam.c index ac187aa..d5f22cd 100644 --- a/libteam/libteam.c +++ b/libteam/libteam.c @@ -236,6 +236,10 @@ int check_call_change_handlers(struct team_handle *th, break; } } + if (call_type_mask & TEAM_IFINFO_CHANGE) { + ifinfo_destroy_removed(th); + ifinfo_clear_changed(th); + } th->change_handler.pending_type_mask &= ~call_type_mask; return err; } @@ -546,6 +550,11 @@ int team_destroy(struct team_handle *th) #endif /* \endcond */
+/* libnl uses default 32k socket receive buffer size, + * whicn can get too small. Use 96k for all sockets. + */ +#define NETLINK_RCVBUF 98304 + /** * @param th libteam library context * @param ifindex team device interface index @@ -561,6 +570,8 @@ int team_init(struct team_handle *th, uint32_t ifindex) int err; int grp_id; int val; + int eventbufsize; + const char *env;
if (!ifindex) { err(th, "Passed interface index %d is not valid.", ifindex); @@ -589,12 +600,12 @@ int team_init(struct team_handle *th, uint32_t ifindex) return -errno; }
- err = nl_socket_set_buffer_size(th->nl_sock, 98304, 0); + err = nl_socket_set_buffer_size(th->nl_sock, NETLINK_RCVBUF, 0); if (err) { err(th, "Failed to set buffer size of netlink sock."); return -nl2syserr(err); } - err = nl_socket_set_buffer_size(th->nl_sock_event, 98304, 0); + err = nl_socket_set_buffer_size(th->nl_sock_event, NETLINK_RCVBUF, 0); if (err) { err(th, "Failed to set buffer size of netlink event sock."); return -nl2syserr(err); @@ -627,6 +638,25 @@ int team_init(struct team_handle *th, uint32_t ifindex) nl_socket_modify_cb(th->nl_cli.sock_event, NL_CB_VALID, NL_CB_CUSTOM, cli_event_handler, th); nl_cli_connect(th->nl_cli.sock_event, NETLINK_ROUTE); + + env = getenv("TEAM_EVENT_BUFSIZE"); + if (env) { + eventbufsize = strtol(env, NULL, 10); + /* ignore other errors, libnl forces minimum 32k and + * too large values are truncated to system rmem_max + */ + if (eventbufsize < 0) + eventbufsize = 0; + } else { + eventbufsize = NETLINK_RCVBUF; + } + + err = nl_socket_set_buffer_size(th->nl_cli.sock_event, eventbufsize, 0); + if (err) { + err(th, "Failed to set cli event socket buffer size."); + return err; + } + err = nl_socket_add_membership(th->nl_cli.sock_event, RTNLGRP_LINK); if (err < 0) { err(th, "Failed to add netlink membership."); @@ -767,7 +797,23 @@ static int get_cli_sock_event_fd(struct team_handle *th)
static int cli_sock_event_handler(struct team_handle *th) { - nl_recvmsgs_default(th->nl_cli.sock_event); + int err; + + err = nl_recvmsgs_default(th->nl_cli.sock_event); + err = -nl2syserr(err); + + /* libnl thinks ENOBUFS and ENOMEM are same. Hope it was ENOBUFS. */ + if (err == -ENOMEM) { + warn(th, "Lost link notifications from kernel."); + /* There's no way to know what events were lost and no + * way to get them again. Refresh all. + */ + err = get_ifinfo_list(th); + } + + if (err) + return err; + return check_call_change_handlers(th, TEAM_IFINFO_CHANGE); }
diff --git a/libteam/team_private.h b/libteam/team_private.h index a07632f..a5eb0be 100644 --- a/libteam/team_private.h +++ b/libteam/team_private.h @@ -115,6 +115,9 @@ int ifinfo_link_with_port(struct team_handle *th, uint32_t ifindex, int ifinfo_link(struct team_handle *th, uint32_t ifindex, struct team_ifinfo **p_ifinfo); void ifinfo_unlink(struct team_ifinfo *ifinfo); +void ifinfo_clear_changed(struct team_handle *th); +void ifinfo_destroy_removed(struct team_handle *th); +int get_ifinfo_list(struct team_handle *th); int get_options_handler(struct nl_msg *msg, void *arg); int option_list_alloc(struct team_handle *th); int option_list_init(struct team_handle *th);
Mon, Feb 06, 2017 at 02:41:05PM CET, atiainen@forcepoint.com wrote:
When there's a large number of interfaces (e.g. vlans), teamd loses link notifications as it cannot read them as fast as kernel is broadcasting them. This often prevents teamd starting properly if started concurrently when other links are being set up. It can also fail when it's up and running, especially in the cases where the team device itself has a lot of vlans under it.
This can easily be reproduces by simple example (in SMP system) by manually adding team device with a bunch of vlans, putting it up, and starting teamd with --take-over option:
root@debian:~# ip link add name team0 type team root@debian:~# for i in `seq 100 150` ; do
ip link add link team0 name team0.$i type vlan id $i ; done
root@debian:~# ip link set team0 up root@debian:~# cat teamd.conf { "device": "team0", "runner": { "name": "activebackup" }, "ports": { "eth1": {}, "eth2": {} } } root@debian:~# teamd -o -N -f teamd.conf
At this point, teamd will not give any error messages or other indication that something is wrong. But state will not look healthy:
root@debian:~# teamdctl team0 state setup: runner: activebackup ports: eth1 link watches: link summary: up instance[link_watch_0]: name: ethtool link: up down count: 0 Failed to parse JSON port dump. command call failed (Invalid argument)
If checking state dump, it will show that port eth2 is missing info. Running strace to teamd will reveal that there's one recvmsgs() that returned -1 with errno ENOBUFS. What happened in this example was that when teamd started, all vlans got carrier up, and kernel flooded notifications faster than teamd could read them. It then lost events related to port eth2 getting enslaved and up.
The socket that joins RTNLGRP_LINK notifications uses default libnl 32k buffer size. Netlink messages are large (over 1k), and this buffer gets easily full. Kernel neither knows nor cares were notification broadcasts delivered. This cannot be fixed by simply increasing the buffer size, as there's no size that is guaranteed to work in every use case, and this can require several megabytes of buffer (a way over normal rmem_max limit) if there are hunderds of vlans.
Only way to recover from this is to refresh all ifinfo list, as it's invalidated at this point. It cannot easily work around of this by just refreshing team device and its ports, because library side might not have ports linked due to events missed, and it doesn't know about teamd configuration.
Checks now return value of nl_recvmsgs_default() for event socket. In case of ENOBUFS (which libnl nicely changes to ENOMEM), refreshes all ifinfo list. get_ifinfo_list() also checks now for removed interfaces in case of missed dellink event. Currently all TEAM_IFINFO_CHANGE handlers processed events one by one, so it had to be changed to support multiple ifinfo changes. For this, ifinfo changed flags are cleared and removed entries destroyed only after all handlers have been called.
Also, increased nl_cli.sock_event receive buffers to 96k like all other sockets. Added possibility to change this via environment variable.
Signed-off-by: Antti Tiainen atiainen@forcepoint.com
applied, thanks!
libteam@lists.fedorahosted.org
-
Antti Tiainen -
Jiri Pirko