kernel_optimize_test/net/core
Tejun Heo a05d4fd917 cgroup, net_cls: iterate the fds of only the tasks which are being migrated
The net_cls controller controls the classid field of each socket which
is associated with the cgroup.  Because the classid is per-socket
attribute, when a task migrates to another cgroup or the configured
classid of the cgroup changes, the controller needs to walk all
sockets and update the classid value, which was implemented by
3b13758f51 ("cgroups: Allow dynamically changing net_classid").

While the approach is not scalable, migrating tasks which have a lot
of fds attached to them is rare and the cost is born by the ones
initiating the operations.  However, for simplicity, both the
migration and classid config change paths call update_classid() which
scans all fds of all tasks in the target css.  This is an overkill for
the migration path which only needs to cover a much smaller subset of
tasks which are actually getting migrated in.

On cgroup v1, this can lead to unexpected scalability issues when one
tries to migrate a task or process into a net_cls cgroup which already
contains a lot of fds.  Even if the migration traget doesn't have many
to get scanned, update_classid() ends up scanning all fds in the
target cgroup which can be extremely numerous.

Unfortunately, on cgroup v2 which doesn't use net_cls, the problem is
even worse.  Before bfc2cf6f61 ("cgroup: call subsys->*attach() only
for subsystems which are actually affected by migration"), cgroup core
would call the ->css_attach callback even for controllers which don't
see actual migration to a different css.

As net_cls is always disabled but still mounted on cgroup v2, whenever
a process is migrated on the cgroup v2 hierarchy, net_cls sees
identity migration from root to root and cgroup core used to call
->css_attach callback for those.  The net_cls ->css_attach ends up
calling update_classid() on the root net_cls css to which all
processes on the system belong to as the controller isn't used.  This
makes any cgroup v2 migration O(total_number_of_fds_on_the_system)
which is horrible and easily leads to noticeable stalls triggering RCU
stall warnings and so on.

The worst symptom is already fixed in upstream by bfc2cf6f61
("cgroup: call subsys->*attach() only for subsystems which are
actually affected by migration"); however, backporting that commit is
too invasive and we want to avoid other cases too.

This patch updates net_cls's cgrp_attach() to iterate fds of only the
processes which are actually getting migrated.  This removes the
surprising migration cost which is dependent on the total number of
fds in the target cgroup.  As this leaves write_classid() the only
user of update_classid(), open-code the helper into write_classid().

Reported-by: David Goode <dgoode@fb.com>
Fixes: 3b13758f51 ("cgroups: Allow dynamically changing net_classid")
Cc: stable@vger.kernel.org # v4.4+
Cc: Nina Schiff <ninasc@fb.com>
Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-03-22 10:32:46 -07:00
..
datagram.c udp: properly cope with csum errors 2017-02-07 11:19:00 -05:00
dev_addr_lists.c
dev_ioctl.c
dev.c net: Resend IGMP memberships upon peer notification. 2017-03-14 11:33:44 -07:00
devlink.c devlink: allow to fillup eswitch attrs even if mode_get op does not exist 2017-02-10 14:43:00 -05:00
drop_monitor.c drop_monitor: consider inserted data in genlmsg_end 2017-01-03 11:09:44 -05:00
dst_cache.c net: dst_cache_per_cpu_dst_set() can be static 2016-03-18 17:45:08 -04:00
dst.c net: pending_confirm is not used anymore 2017-02-07 13:07:47 -05:00
ethtool.c sched/headers: Prepare to move signal wakeup & sigpending methods from <linux/sched.h> into <linux/sched/signal.h> 2017-03-02 08:42:32 +01:00
fib_rules.c net: core: add missing check for uid_range in rule_exists. 2016-11-09 13:28:10 -05:00
filter.c bpf: Fix bpf_xdp_event_output 2017-02-23 13:53:42 -05:00
flow_dissector.c flow dissector: check if arp_eth is null rather than arp 2017-01-16 13:48:48 -05:00
flow.c Merge branch 'smp-hotplug-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip 2016-12-12 19:25:04 -08:00
gen_estimator.c Replace <asm/uaccess.h> with <linux/uaccess.h> globally 2016-12-24 11:46:01 -08:00
gen_stats.c net_sched: gen_estimator: complete rewrite of rate estimators 2016-12-05 15:21:59 -05:00
gro_cells.c gro_cells: move to net/core/gro_cells.c 2017-02-08 14:38:18 -05:00
hwbm.c net: hwbm: Fix unbalanced spinlock in error case 2016-05-25 12:35:09 -07:00
link_watch.c dev: introduce dev_get_iflink() 2015-04-02 14:04:59 -04:00
lwt_bpf.c lwtunnel: remove device arg to lwtunnel_build_state 2017-01-30 15:14:22 -05:00
lwtunnel.c lwtunnel: remove device arg to lwtunnel_build_state 2017-01-30 15:14:22 -05:00
Makefile gro_cells: move to net/core/gro_cells.c 2017-02-08 14:38:18 -05:00
neighbour.c net: neigh: Fix netevent NETEVENT_DELAY_PROBE_TIME_UPDATE notification 2017-02-15 12:38:43 -05:00
net_namespace.c sched/headers: Prepare to move the task_lock()/unlock() APIs to <linux/sched/task.h> 2017-03-02 08:42:38 +01:00
net-procfs.c net: remove NETDEV_TX_LOCKED support 2016-04-26 15:53:05 -04:00
net-sysfs.c net: use net->count to check whether a netns is alive or not 2017-03-13 16:02:27 -07:00
net-sysfs.h
net-traces.c net: IPv6 fib lookup tracepoint 2015-11-22 11:54:10 -05:00
netclassid_cgroup.c cgroup, net_cls: iterate the fds of only the tasks which are being migrated 2017-03-22 10:32:46 -07:00
netevent.c netevent: remove automatic variable in register_netevent_notifier() 2015-05-31 00:03:21 -07:00
netpoll.c netpoll: more efficient locking 2016-11-16 18:32:02 -05:00
netprio_cgroup.c sched/headers: Prepare to move the task_lock()/unlock() APIs to <linux/sched/task.h> 2017-03-02 08:42:38 +01:00
pktgen.c net-tc: convert tc_verd to integer bitfields 2017-01-08 20:58:52 -05:00
ptp_classifier.c ptp: Change ptp_class to a proper bitmask 2015-11-03 11:08:22 -05:00
request_sock.c ipv4: Namespaceify tcp_max_syn_backlog knob 2016-12-29 11:38:31 -05:00
rtnetlink.c rtnl: simplify error return path in rtnl_create_link() 2017-02-21 12:17:43 -05:00
scm.c sched/headers: Prepare for new header dependencies before moving code to <linux/sched/user.h> 2017-03-02 08:42:29 +01:00
secure_seq.c secure_seq: fix sparse errors 2017-01-12 15:57:10 -05:00
skbuff.c tcp: mark skbs with SCM_TIMESTAMPING_OPT_STATS 2017-03-21 18:44:17 -07:00
sock_diag.c sock_diag: align nlattr properly when needed 2016-04-26 12:00:48 -04:00
sock_reuseport.c soreuseport: do not export reuseport_add_sock() 2016-10-18 14:18:23 -04:00
sock.c net: properly release sk_frag.page 2017-03-15 15:37:45 -07:00
stream.c sched/headers: Prepare for new header dependencies before moving code to <linux/sched/signal.h> 2017-03-02 08:42:29 +01:00
sysctl_net_core.c bpf: make jited programs visible in traces 2017-02-17 13:40:05 -05:00
timestamping.c net: skb_defer_rx_timestamp should check for phydev before setting up classify 2015-07-09 14:17:15 -07:00
tso.c net: tso: add support for IPv6 2015-10-26 22:24:22 -07:00
utils.c Replace <asm/uaccess.h> with <linux/uaccess.h> globally 2016-12-24 11:46:01 -08:00