linux-stable.git/net/sched, branch linux-4.0.y

net_sched: invoke ->attach() after setting dev->qdisc

2015-06-23T00:03:22+00:00

[ Upstream commit 86e363dc3b50bfd50a1f315934583fbda673ab8d ]

For mq qdisc, we add per tx queue qdisc to root qdisc
for display purpose, however, that happens too early,
before the new dev->qdisc is finally set, this causes
q->list points to an old root qdisc which is going to be
freed right before assigning with a new one.

Fix this by moving ->attach() after setting dev->qdisc.

For the record, this fixes the following crash:

 ------------[ cut here ]------------
 WARNING: CPU: 1 PID: 975 at lib/list_debug.c:59 __list_del_entry+0x5a/0x98()
 list_del corruption. prev->next should be ffff8800d1998ae8, but was 6b6b6b6b6b6b6b6b
 CPU: 1 PID: 975 Comm: tc Not tainted 4.1.0-rc4+ #1019
 Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011
  0000000000000009 ffff8800d73fb928 ffffffff81a44e7f 0000000047574756
  ffff8800d73fb978 ffff8800d73fb968 ffffffff810790da ffff8800cfc4cd20
  ffffffff814e725b ffff8800d1998ae8 ffffffff82381250 0000000000000000
 Call Trace:
  [] dump_stack+0x4c/0x65
  [] warn_slowpath_common+0x9c/0xb6
  [] ? __list_del_entry+0x5a/0x98
  [] warn_slowpath_fmt+0x46/0x48
  [] ? dev_graft_qdisc+0x5e/0x6a
  [] __list_del_entry+0x5a/0x98
  [] list_del+0xe/0x2d
  [] qdisc_list_del+0x1e/0x20
  [] qdisc_destroy+0x30/0xd6
  [] qdisc_graft+0x11d/0x243
  [] tc_get_qdisc+0x1a6/0x1d4
  [] ? mark_lock+0x2e/0x226
  [] rtnetlink_rcv_msg+0x181/0x194
  [] ? rtnl_lock+0x17/0x19
  [] ? rtnl_lock+0x17/0x19
  [] ? __rtnl_unlock+0x17/0x17
  [] netlink_rcv_skb+0x4d/0x93
  [] rtnetlink_rcv+0x26/0x2d
  [] netlink_unicast+0xcb/0x150
  [] ? might_fault+0x59/0xa9
  [] netlink_sendmsg+0x4fa/0x51c
  [] sock_sendmsg_nosec+0x12/0x1d
  [] sock_sendmsg+0x29/0x2e
  [] ___sys_sendmsg+0x1b4/0x23a
  [] ? native_sched_clock+0x35/0x37
  [] ? sched_clock_local+0x12/0x72
  [] ? sched_clock_cpu+0x9e/0xb7
  [] ? current_kernel_time+0xe/0x32
  [] ? lock_release_holdtime.part.29+0x71/0x7f
  [] ? read_seqcount_begin.constprop.27+0x5f/0x76
  [] ? trace_hardirqs_on_caller+0x17d/0x199
  [] ? __fget_light+0x50/0x78
  [] __sys_sendmsg+0x42/0x60
  [] SyS_sendmsg+0x12/0x1c
  [] system_call_fastpath+0x12/0x6f
 ---[ end trace ef29d3fb28e97ae7 ]---

For long term, we probably need to clean up the qdisc_graft() code
in case it hides other bugs like this.

Fixes: 95dc19299f74 ("pkt_sched: give visibility to mq slave qdiscs")
Cc: Jamal Hadi Salim 
Signed-off-by: Cong Wang 
Acked-by: Eric Dumazet 
Signed-off-by: David S. Miller 
Signed-off-by: Greg Kroah-Hartman

net: sched: fix call_rcu() race on classifier module unloads

2015-06-23T00:03:20+00:00

[ Upstream commit c78e1746d3ad7d548bdf3fe491898cc453911a49 ]

Vijay reported that a loop as simple as ...

  while true; do
    tc qdisc add dev foo root handle 1: prio
    tc filter add dev foo parent 1: u32 match u32 0 0  flowid 1
    tc qdisc del dev foo root
    rmmod cls_u32
  done

... will panic the kernel. Moreover, he bisected the change
apparently introducing it to 78fd1d0ab072 ("netlink: Re-add
locking to netlink_lookup() and seq walker").

The removal of synchronize_net() from the netlink socket
triggering the qdisc to be removed, seems to have uncovered
an RCU resp. module reference count race from the tc API.
Given that RCU conversion was done after e341694e3eb5 ("netlink:
Convert netlink_lookup() to use RCU protected hash table")
which added the synchronize_net() originally, occasion of
hitting the bug was less likely (not impossible though):

When qdiscs that i) support attaching classifiers and,
ii) have at least one of them attached, get deleted, they
invoke tcf_destroy_chain(), and thus call into ->destroy()
handler from a classifier module.

After RCU conversion, all classifier that have an internal
prio list, unlink them and initiate freeing via call_rcu()
deferral.

Meanhile, tcf_destroy() releases already reference to the
tp->ops->owner module before the queued RCU callback handler
has been invoked.

Subsequent rmmod on the classifier module is then not prevented
since all module references are already dropped.

By the time, the kernel invokes the RCU callback handler from
the module, that function address is then invalid.

One way to fix it would be to add an rcu_barrier() to
unregister_tcf_proto_ops() to wait for all pending call_rcu()s
to complete.

synchronize_rcu() is not appropriate as under heavy RCU
callback load, registered call_rcu()s could be deferred
longer than a grace period. In case we don't have any pending
call_rcu()s, the barrier is allowed to return immediately.

Since we came here via unregister_tcf_proto_ops(), there
are no users of a given classifier anymore. Further nested
call_rcu()s pointing into the module space are not being
done anywhere.

Only cls_bpf_delete_prog() may schedule a work item, to
unlock pages eventually, but that is not in the range/context
of cls_bpf anymore.

Fixes: 25d8c0d55f24 ("net: rcu-ify tcf_proto")
Fixes: 9888faefe132 ("net: sched: cls_basic use RCU")
Reported-by: Vijay Subramanian 
Signed-off-by: Daniel Borkmann 
Cc: John Fastabend 
Cc: Eric Dumazet 
Cc: Thomas Graf 
Cc: Jamal Hadi Salim 
Cc: Alexei Starovoitov 
Tested-by: Vijay Subramanian 
Acked-by: Alexei Starovoitov 
Acked-by: Eric Dumazet 
Signed-off-by: David S. Miller 
Signed-off-by: Greg Kroah-Hartman

act_bpf: allow non-default TC_ACT opcodes as BPF exec outcome

2015-03-18T02:15:06+00:00

Revisiting commit d23b8ad8ab23 ("tc: add BPF based action") with regards
to eBPF support, I was thinking that it might be better to improve
return semantics from a BPF program invoked through BPF_PROG_RUN().

Currently, in case filter_res is 0, we overwrite the default action
opcode with TC_ACT_SHOT. A default action opcode configured through tc's
m_bpf can be: TC_ACT_RECLASSIFY, TC_ACT_PIPE, TC_ACT_SHOT, TC_ACT_UNSPEC,
TC_ACT_OK.

In cls_bpf, we have the possibility to overwrite the default class
associated with the classifier in case filter_res is _not_ 0xffffffff
(-1).

That allows us to fold multiple [e]BPF programs into a single one, where
they would otherwise need to be defined as a separate classifier with
its own classid, needlessly redoing parsing work, etc.

Similarly, we could do better in act_bpf: Since above TC_ACT* opcodes
are exported to UAPI anyway, we reuse them for return-code-to-tc-opcode
mapping, where we would allow above possibilities. Thus, like in cls_bpf,
a filter_res of 0xffffffff (-1) means that the configured _default_ action
is used. Any unkown return code from the BPF program would fail in
tcf_bpf() with TC_ACT_UNSPEC.

Should we one day want to make use of TC_ACT_STOLEN or TC_ACT_QUEUED,
which both have the same semantics, we have the option to either use
that as a default action (filter_res of 0xffffffff) or non-default BPF
return code.

All that will allow us to transparently use tcf_bpf() for both BPF
flavours.

Signed-off-by: Daniel Borkmann 
Cc: Jiri Pirko 
Cc: Alexei Starovoitov 
Cc: Jamal Hadi Salim 
Acked-by: Jiri Pirko 
Signed-off-by: David S. Miller

net_sched: fix struct tc_u_hnode layout in u32

2015-03-10T03:44:31+00:00

We dynamically allocate divisor+1 entries for ->ht[] in tc_u_hnode:

  ht = kzalloc(sizeof(*ht) + divisor*sizeof(void *), GFP_KERNEL);

So ->ht is supposed to be the last field of this struct, however
this is broken, since an rcu head is appended after it.

Fixes: 1ce87720d456 ("net: sched: make cls_u32 lockless")
Cc: Jamal Hadi Salim 
Cc: John Fastabend 
Signed-off-by: Cong Wang 
Acked-by: Eric Dumazet 
Signed-off-by: David S. Miller

Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net

2015-03-03T23:30:07+00:00

Pull networking fixes from David Miller:

 1) If an IPVS tunnel is created with a mixed-family destination
    address, it cannot be removed.  Fix from Alexey Andriyanov.

 2) Fix module refcount underflow in netfilter's nft_compat, from Pablo
    Neira Ayuso.

 3) Generic statistics infrastructure can reference variables sitting on
    a released function stack, therefore use dynamic allocation always.
    Fix from Ignacy Gawędzki.

 4) skb_copy_bits() return value test is inverted in ip_check_defrag().

 5) Fix network namespace exit in openvswitch, we have to release all of
    the per-net vports.  From Pravin B Shelar.

 6) Fix signedness bug in CAIF's cfpkt_iterate(), from Dan Carpenter.

 7) Fix rhashtable grow/shrink behavior, only expand during inserts and
    shrink during deletes.  From Daniel Borkmann.

 8) Netdevice names with semicolons should never be allowed, because
    they serve as a separator.  From Matthew Thode.

 9) Use {,__}set_current_state() where appropriate, from Fabian
    Frederick.

10) Revert byte queue limits support in r8169 driver, it's causing
    regressions we can't figure out.

11) tcp_should_expand_sndbuf() erroneously uses tp->packets_out to
    measure packets in flight, properly use tcp_packets_in_flight()
    instead.  From Neal Cardwell.

12) Fix accidental removal of support for bluetooth in CSR based Intel
    wireless cards.  From Marcel Holtmann.

13) We accidently added a behavioral change between native and compat
    tasks, wrt testing the MSG_CMSG_COMPAT bit.  Just ignore it if the
    user happened to set it in a native binary as that was always the
    behavior we had.  From Catalin Marinas.

14) Check genlmsg_unicast() return valud in hwsim netlink tx frame
    handling, from Bob Copeland.

15) Fix stale ->radar_required setting in mac80211 that can prevent
    starting new scans, from Eliad Peller.

16) Fix memory leak in nl80211 monitor, from Johannes Berg.

17) Fix race in TX index handling in xen-netback, from David Vrabel.

18) Don't enable interrupts in amx-xgbe driver until all software et al.
    state is ready for the interrupt handler to run.  From Thomas
    Lendacky.

19) Add missing netlink_ns_capable() checks to rtnl_newlink(), from Eric
    W Biederman.

20) The amount of header space needed in macvtap was not calculated
    properly, fix it otherwise we splat past the beginning of the
    packet.  From Eric Dumazet.

21) Fix bcmgenet TCP TX perf regression, from Jaedon Shin.

22) Don't raw initialize or mod timers, use setup_timer() and
    mod_timer() instead.  From Vaishali Thakkar.

23) Fix software maintained statistics in bcmgenet and systemport
    drivers, from Florian Fainelli.

24) DMA descriptor updates in sh_eth need proper memory barriers, from
    Ben Hutchings.

25) Don't do UDP Fragmentation Offload on RAW sockets, from Michal
    Kubecek.

26) Openvswitch's non-masked set actions aren't constructed properly
    into netlink messages, fix from Joe Stringer.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (116 commits)
  openvswitch: Fix serialization of non-masked set actions.
  gianfar: Reduce logging noise seen due to phy polling if link is down
  ibmveth: Add function to enable live MAC address changes
  net: bridge: add compile-time assert for cb struct size
  udp: only allow UFO for packets from SOCK_DGRAM sockets
  sh_eth: Really fix padding of short frames on TX
  Revert "sh_eth: Enable Rx descriptor word 0 shift for r8a7790"
  sh_eth: Fix RX recovery on R-Car in case of RX ring underrun
  sh_eth: Ensure proper ordering of descriptor active bit write/read
  net/mlx4_en: Disbale GRO for incoming loopback/selftest packets
  net/mlx4_core: Fix wrong mask and error flow for the update-qp command
  net: systemport: fix software maintained statistics
  net: bcmgenet: fix software maintained statistics
  rxrpc: don't multiply with HZ twice
  rxrpc: terminate retrans loop when sending of skb fails
  net/hsr: Fix NULL pointer dereference and refcnt bugs when deleting a HSR interface.
  net: pasemi: Use setup_timer and mod_timer
  net: stmmac: Use setup_timer and mod_timer
  net: 8390: axnet_cs: Use setup_timer and mod_timer
  net: 8390: pcnet_cs: Use setup_timer and mod_timer
  ...

ematch: Fix auto-loading of ematch modules.

2015-02-20T20:30:56+00:00

In tcf_em_validate(), after calling request_module() to load the
kind-specific module, set em->ops to NULL before returning -EAGAIN, so
that module_put() is not called again by tcf_em_tree_destroy().

Signed-off-by: Ignacy Gawędzki 
Acked-by: Cong Wang 
Signed-off-by: David S. Miller

Merge branch 'kconfig' of git://git.kernel.org/pub/scm/linux/kernel/git/mmarek/kbuild

2015-02-19T18:36:45+00:00

Pull kconfig updates from Michal Marek:
 "Yann E Morin was supposed to take over kconfig maintainership, but
  this hasn't happened.  So I'm sending a few kconfig patches that I
  collected:

   - Fix for missing va_end in kconfig
   - merge_config.sh displays used if given too few arguments
   - s/boolean/bool/ in Kconfig files for consistency, with the plan to
     only support bool in the future"

* 'kconfig' of git://git.kernel.org/pub/scm/linux/kernel/git/mmarek/kbuild:
  kconfig: use va_end to match corresponding va_start
  merge_config.sh: Display usage if given too few arguments
  kconfig: use bool instead of boolean for type definition attributes

Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net

2015-02-05T22:33:28+00:00

Conflicts:
	drivers/net/vxlan.c
	drivers/vhost/net.c
	include/linux/if_vlan.h
	net/core/dev.c

The net/core/dev.c conflict was the overlap of one commit marking an
existing function static whilst another was adding a new function.

In the include/linux/if_vlan.h case, the type used for a local
variable was changed in 'net', whereas the function got rewritten
to fix a stacked vlan bug in 'net-next'.

In drivers/vhost/net.c, Al Viro's iov_iter conversions in 'net-next'
overlapped with an endainness fix for VHOST 1.0 in 'net'.

In drivers/net/vxlan.c, vxlan_find_vni() added a 'flags' parameter
in 'net-next' whereas in 'net' there was a bug fix to pass in the
correct network namespace pointer in calls to this function.

Signed-off-by: David S. Miller

pkt_sched: fq: better control of DDOS traffic

2015-02-05T06:15:45+00:00

FQ has a fast path for skb attached to a socket, as it does not
have to compute a flow hash. But for other packets, FQ being non
stochastic means that hosts exposed to random Internet traffic
can allocate million of flows structure (104 bytes each) pretty
easily. Not only host can OOM, but lookup in RB trees can take
too much cpu and memory resources.

This patch adds a new attribute, orphan_mask, that is adding
possibility of having a stochastic hash for orphaned skb.

Its default value is 1024 slots, to mimic SFQ behavior.

Note: This does not apply to locally generated TCP traffic,
and no locally generated traffic will share a flow structure
with another perfect or stochastic flow.

This patch also handles the specific case of SYNACK messages:

They are attached to the listener socket, and therefore all map
to a single hash bucket. If listener have set SO_MAX_PACING_RATE,
hoping to have new accepted socket inherit this rate, SYNACK
might be paced and even dropped.

This is very similar to an internal patch Google have used more
than one year.

Signed-off-by: Eric Dumazet 
Signed-off-by: David S. Miller

tcp: do not pace pure ack packets

2015-02-05T04:36:31+00:00

When we added pacing to TCP, we decided to let sch_fq take care
of actual pacing.

All TCP had to do was to compute sk->pacing_rate using simple formula:

sk->pacing_rate = 2 * cwnd * mss / rtt

It works well for senders (bulk flows), but not very well for receivers
or even RPC :

cwnd on the receiver can be less than 10, rtt can be around 100ms, so we
can end up pacing ACK packets, slowing down the sender.

Really, only the sender should pace, according to its own logic.

Instead of adding a new bit in skb, or call yet another flow
dissection, we tweak skb->truesize to a small value (2), and
we instruct sch_fq to use new helper and not pace pure ack.

Note this also helps TCP small queue, as ack packets present
in qdisc/NIC do not prevent sending a data packet (RPC workload)

This helps to reduce tx completion overhead, ack packets can use regular
sock_wfree() instead of tcp_wfree() which is a bit more expensive.

This has no impact in the case packets are sent to loopback interface,
as we do not coalesce ack packets (were we would detect skb->truesize
lie)

In case netem (with a delay) is used, skb_orphan_partial() also sets
skb->truesize to 1.

This patch is a combination of two patches we used for about one year at
Google.

Signed-off-by: Eric Dumazet 
Signed-off-by: David S. Miller