linux.git/net/core, branch v5.6-rc7

net: memcg: late association of sock to memcg

2020-03-10T22:33:05+00:00

If a TCP socket is allocated in IRQ context or cloned from unassociated
(i.e. not associated to a memcg) in IRQ context then it will remain
unassociated for its whole life. Almost half of the TCPs created on the
system are created in IRQ context, so, memory used by such sockets will
not be accounted by the memcg.

This issue is more widespread in cgroup v1 where network memory
accounting is opt-in but it can happen in cgroup v2 if the source socket
for the cloning was created in root memcg.

To fix the issue, just do the association of the sockets at the accept()
time in the process context and then force charge the memory buffer
already used and reserved by the socket.

Signed-off-by: Shakeel Butt 
Reviewed-by: Eric Dumazet 
Signed-off-by: David S. Miller

cgroup, netclassid: periodically release file_lock on classid updating

2020-03-10T01:13:39+00:00

In our production environment we have faced with problem that updating
classid in cgroup with heavy tasks cause long freeze of the file tables
in this tasks. By heavy tasks we understand tasks with many threads and
opened sockets (e.g. balancers). This freeze leads to an increase number
of client timeouts.

This patch implements following logic to fix this issue:
аfter iterating 1000 file descriptors file table lock will be released
thus providing a time gap for socket creation/deletion.

Now update is non atomic and socket may be skipped using calls:

dup2(oldfd, newfd);
close(oldfd);

But this case is not typical. Moreover before this patch skip is possible
too by hiding socket fd in unix socket buffer.

New sockets will be allocated with updated classid because cgroup state
is updated before start of the file descriptors iteration.

So in common cases this patch has no side effects.

Signed-off-by: Dmitry Yakunin 
Reviewed-by: Konstantin Khlebnikov 
Signed-off-by: David S. Miller

devlink: validate length of region addr/len

2020-03-03T21:28:48+00:00

DEVLINK_ATTR_REGION_CHUNK_ADDR and DEVLINK_ATTR_REGION_CHUNK_LEN
lack entries in the netlink policy. Corresponding nla_get_u64()s
may read beyond the end of the message.

Fixes: 4e54795a27f5 ("devlink: Add support for region snapshot read command")
Signed-off-by: Jakub Kicinski 
Reviewed-by: Jiri Pirko 
Signed-off-by: David S. Miller

devlink: validate length of param values

2020-03-03T21:28:48+00:00

DEVLINK_ATTR_PARAM_VALUE_DATA may have different types
so it's not checked by the normal netlink policy. Make
sure the attribute length is what we expect.

Fixes: e3b7ca18ad7b ("devlink: Add param set command")
Signed-off-by: Jakub Kicinski 
Reviewed-by: Jiri Pirko 
Signed-off-by: David S. Miller

net: core: devlink.c: Use built-in RCU list checking

2020-02-27T00:59:18+00:00

list_for_each_entry_rcu() has built-in RCU and lock checking.

Pass cond argument to list_for_each_entry_rcu() to silence
false lockdep warning when CONFIG_PROVE_RCU_LIST is enabled.

The devlink->lock is held when devlink_dpipe_table_find()
is called in non RCU read side section. Therefore, pass struct devlink
to devlink_dpipe_table_find() for lockdep checking.

Signed-off-by: Madhuparna Bhowmik 
Reviewed-by: Jiri Pirko 
Signed-off-by: David S. Miller

net: Fix Tx hash bound checking

2020-02-26T19:14:10+00:00

Fixes the lower and upper bounds when there are multiple TCs and
traffic is on the the same TC on the same device.

The lower bound is represented by 'qoffset' and the upper limit for
hash value is 'qcount + qoffset'. This gives a clean Rx to Tx queue
mapping when there are multiple TCs, as the queue indices for upper TCs
will be offset by 'qoffset'.

v2: Fixed commit description based on comments.

Fixes: 1b837d489e06 ("net: Revoke export for __skb_tx_hash, update it to just be static skb_tx_hash")
Fixes: eadec877ce9c ("net: Add support for subordinate traffic classes to netdev_pick_tx")
Signed-off-by: Amritha Nambiar 
Reviewed-by: Alexander Duyck 
Reviewed-by: Sridhar Samudrala 
Signed-off-by: David S. Miller

net: core: devlink.c: Hold devlink->lock from the beginning of devlink_dpipe_table_register()

2020-02-24T05:17:37+00:00

devlink_dpipe_table_find() should be called under either
rcu_read_lock() or devlink->lock. devlink_dpipe_table_register()
calls devlink_dpipe_table_find() without holding the lock
and acquires it later. Therefore hold the devlink->lock
from the beginning of devlink_dpipe_table_register().

Suggested-by: Jiri Pirko 
Signed-off-by: Madhuparna Bhowmik 
Reviewed-by: Jiri Pirko 
Signed-off-by: David S. Miller

net: core: Distribute switch variables for initialization

2020-02-20T18:00:19+00:00

Variables declared in a switch statement before any case statements
cannot be automatically initialized with compiler instrumentation (as
they are not part of any execution flow). With GCC's proposed automatic
stack variable initialization feature, this triggers a warning (and they
don't get initialized). Clang's automatic stack variable initialization
(via CONFIG_INIT_STACK_ALL=y) doesn't throw a warning, but it also
doesn't initialize such variables[1]. Note that these warnings (or silent
skipping) happen before the dead-store elimination optimization phase,
so even when the automatic initializations are later elided in favor of
direct initializations, the warnings remain.

To avoid these problems, move such variables into the "case" where
they're used or lift them up into the main function body.

net/core/skbuff.c: In function ‘skb_checksum_setup_ip’:
net/core/skbuff.c:4809:7: warning: statement will never be executed [-Wswitch-unreachable]
 4809 |   int err;
      |       ^~~

[1] https://bugs.llvm.org/show_bug.cgi?id=44916

Signed-off-by: Kees Cook 
Signed-off-by: David S. Miller

Revert "net: dev: introduce support for sch BYPASS for lockless qdisc"

2020-02-18T20:37:45+00:00

This reverts commit ba27b4cdaaa66561aaedb2101876e563738d36fe

Ahmed reported ouf-of-order issues bisected to commit ba27b4cdaaa6
("net: dev: introduce support for sch BYPASS for lockless qdisc").
I can't find any working solution other than a plain revert.

This will introduce some minor performance regressions for
pfifo_fast qdisc. I plan to address them in net-next with more
indirect call wrapper boilerplate for qdiscs.

Reported-by: Ahmad Fatoum 
Fixes: ba27b4cdaaa6 ("net: dev: introduce support for sch BYPASS for lockless qdisc")
Signed-off-by: Paolo Abeni 
Signed-off-by: David S. Miller

skbuff: remove stale bit mask comments

2020-02-17T03:49:51+00:00

Remove stale comments since this flag is no longer a bit mask
but is a bit field.

Signed-off-by: Randy Dunlap 
Signed-off-by: David S. Miller