struct bpf_flow_dissector has a small subset of sk_buff fields that
flow dissector BPF program is allowed to access and an optional
pointer to real skb. Real skb is used only in bpf_skb_load_bytes
helper to read non-linear data.
The real motivation for this is to be able to call flow dissector
from eth_get_headlen context where we don't have an skb and need
to dissect raw bytes.
Signed-off-by: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
This revert commit 46b1c18f9d ("net: sched: put back q.qlen into
a single location").
After the previous patch, when a NOLOCK qdisc is enslaved to a
locking qdisc it switches to global stats accounting. As a consequence,
when a classful qdisc accesses directly a child qdisc's qlen, such
qdisc is not doing per CPU accounting and qlen value is consistent.
In the control path nobody uses directly qlen since commit
e5f0e8f8e4 ("net: sched: introduce and use qdisc tree flush/purge
helpers"), so we can remove the contented atomic ops from the
datapath.
v1 -> v2:
- complete the qdisc_qstats_atomic_qlen_dec() ->
qdisc_qstats_cpu_qlen_dec() replacement, fix build issue
- more descriptive commit message
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since stats updating is always consistent with TCQ_F_CPUSTATS flag,
we can disable it at qdisc creation time flipping such bit.
In my experiments, if the NOLOCK flag is cleared, per CPU stats
accounting does not give any measurable performance gain, but it
waste some memory.
Let's clear TCQ_F_CPUSTATS together with NOLOCK, when enslaving
a NOLOCK qdisc to 'lock' one.
Use stats update helper inside pfifo_fast, to cope correctly with
TCQ_F_CPUSTATS flag change.
As a side effect, q.qlen value for any child qdiscs is always
consistent for all lock classfull qdiscs.
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The core sched implementation checks independently for NOLOCK flag
to acquire/release the root spin lock and for qdisc_is_percpu_stats()
to account per CPU values in many places.
This change update the last few places checking the TCQ_F_NOLOCK to
do per CPU stats accounting according to qdisc_is_percpu_stats()
value.
The above allows to clean dev_requeue_skb() implementation a bit
and makes stats update always consistent with a single flag.
v1 -> v2:
- do not move qdisc_is_empty definition, fix build issue
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When checking for root qdisc queue length, do not access directly q.qlen.
In the following patches we will move back qlen accounting to per CPU
values for NOLOCK qdiscs.
Instead, prefer the qdisc_is_empty() helper usage.
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Minor comment merge conflict in mlx5.
Staging driver has a fixup due to the skb->xmit_more changes
in 'net-next', but was removed in 'net'.
Signed-off-by: David S. Miller <davem@davemloft.net>
The same code to flush qdisc tree and purge the qdisc queue
is duplicated in many places and in most cases it does not
respect NOLOCK qdisc: the global backlog len is used and the
per CPU values are ignored.
This change addresses the above, factoring-out the relevant
code and using the helpers introduced by the previous patch
to fetch the correct backlog len.
Fixes: c5ad119fb6 ("net: sched: pfifo_fast use skb_array")
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Classful qdiscs can't access directly the child qdiscs backlog
length: if such qdisc is NOLOCK, per CPU values should be
accounted instead.
Most qdiscs no not respect the above. As a result, qstats fetching
for most classful qdisc is currently incorrect: if the child qdisc is
NOLOCK, it always reports 0 len backlog.
This change introduces a pair of helpers to safely fetch
both backlog and qlen and use them in stats class dumping
functions, fixing the above issue and cleaning a bit the code.
DRR needs also to access the child qdisc queue length, so it
needs custom handling.
Fixes: c5ad119fb6 ("net: sched: pfifo_fast use skb_array")
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The queue is marked not empty after acquiring the seqlock,
and it's up to the NOLOCK qdisc clearing such flag on dequeue.
Since the empty status lays on the same cache-line of the
seqlock, it's always hot on cache during the updates.
This makes the empty flag update a little bit loosy. Given
the lack of synchronization between enqueue and dequeue, this
is unavoidable.
v2 -> v3:
- qdisc_is_empty() has a const argument (Eric)
v1 -> v2:
- use really an 'empty' flag instead of 'not_empty', as
suggested by Eric
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Ivan Vecera <ivecera@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
use RCU when accessing the action chain, to avoid use after free in the
traffic path when 'goto chain' is replaced on existing TC actions (see
script below). Since the control action is read in the traffic path
without holding the action spinlock, we need to explicitly ensure that
a->goto_chain is not NULL before dereferencing (i.e it's not sufficient
to rely on the value of TC_ACT_GOTO_CHAIN bits). Not doing so caused NULL
dereferences in tcf_action_goto_chain_exec() when the following script:
# tc chain add dev dd0 chain 42 ingress protocol ip flower \
> ip_proto udp action pass index 4
# tc filter add dev dd0 ingress protocol ip flower \
> ip_proto udp action csum udp goto chain 42 index 66
# tc chain del dev dd0 chain 42 ingress
(start UDP traffic towards dd0)
# tc action replace action csum udp pass index 66
was run repeatedly for several hours.
Suggested-by: Cong Wang <xiyou.wangcong@gmail.com>
Suggested-by: Vlad Buslov <vladbu@mellanox.com>
Signed-off-by: Davide Caratti <dcaratti@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In the series fc8b81a598 ("Merge branch 'lockless-qdisc-series'")
John made the assumption that the data path had no need to read
the qdisc qlen (number of packets in the qdisc).
It is true when pfifo_fast is used as the root qdisc, or as direct MQ/MQPRIO
children.
But pfifo_fast can be used as leaf in class full qdiscs, and existing
logic needs to access the child qlen in an efficient way.
HTB breaks badly, since it uses cl->leaf.q->q.qlen in :
htb_activate() -> WARN_ON()
htb_dequeue_tree() to decide if a class can be htb_deactivated
when it has no more packets.
HFSC, DRR, CBQ, QFQ have similar issues, and some calls to
qdisc_tree_reduce_backlog() also read q.qlen directly.
Using qdisc_qlen_sum() (which iterates over all possible cpus)
in the data path is a non starter.
It seems we have to put back qlen in a central location,
at least for stable kernels.
For all qdisc but pfifo_fast, qlen is guarded by the qdisc lock,
so the existing q.qlen{++|--} are correct.
For 'lockless' qdisc (pfifo_fast so far), we need to use atomic_{inc|dec}()
because the spinlock might be not held (for example from
pfifo_fast_enqueue() and pfifo_fast_dequeue())
This patch adds atomic_qlen (in the same location than qlen)
and renames the following helpers, since we want to express
they can be used without qdisc lock, and that qlen is no longer percpu.
- qdisc_qstats_cpu_qlen_dec -> qdisc_qstats_atomic_qlen_dec()
- qdisc_qstats_cpu_qlen_inc -> qdisc_qstats_atomic_qlen_inc()
Later (net-next) we might revert this patch by tracking all these
qlen uses and replace them by a more efficient method (not having
to access a precise qlen, but an empty/non_empty status that might
be less expensive to maintain/track).
Another possibility is to have a legacy pfifo_fast version that would
be used when used a a child qdisc, since the parent qdisc needs
a spinlock anyway. But then, future lockless qdiscs would also
have the same problem.
Fixes: 7e66016f2c ("net: sched: helpers to sum qlen and qlen for per cpu logic")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: Cong Wang <xiyou.wangcong@gmail.com>
Cc: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
Extend Qdisc_class_ops with flags. Create enum to hold possible class ops
flag values. Add first class ops flags value QDISC_CLASS_OPS_DOIT_UNLOCKED
to indicate that class ops functions can be called without taking rtnl
lock.
Signed-off-by: Vlad Buslov <vladbu@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add 'rtnl_held' flag to tcf proto change, delete, destroy, dump, walk
functions to track rtnl lock status. Extend users of these function in cls
API to propagate rtnl lock status to them. This allows classifiers to
obtain rtnl lock when necessary and to pass rtnl lock status to extensions
and driver offload callbacks.
Add flags field to tcf proto ops. Add flag value to indicate that
classifier doesn't require rtnl lock.
Signed-off-by: Vlad Buslov <vladbu@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add optional tp->ops->put() API to be implemented for filter reference
counting. This new function is called by cls API to release filter
reference for filters returned by tp->ops->change() or tp->ops->get()
functions. Implement tfilter_put() helper to call tp->ops->put() only for
classifiers that implement it.
Signed-off-by: Vlad Buslov <vladbu@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Extend tcf_chain with 'flushing' flag. Use the flag to prevent insertion of
new classifier instances when chain flushing is in progress in order to
prevent resource leak when tcf_proto is created by unlocked users
concurrently.
Return EAGAIN error from tcf_chain_tp_insert_unique() to restart
tc_new_tfilter() and lookup the chain/proto again.
Signed-off-by: Vlad Buslov <vladbu@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Implement unique insertion function to atomically attach tcf_proto to chain
after verifying that no other tcf proto with specified priority exists.
Implement delete function that verifies that tp is actually empty before
deleting it. Use these functions to refactor cls API to account for
concurrent tp and rule update instead of relying on rtnl lock. Add new
'deleting' flag to tcf proto. Use it to restart search when iterating over
tp's on chain to prevent accessing potentially inval tp->next pointer.
Extend tcf proto with spinlock that is intended to be used to protect its
data from concurrent modification instead of relying on rtnl mutex. Use it
to protect 'deleting' flag. Add lockdep macros to validate that lock is
held when accessing protected fields.
Signed-off-by: Vlad Buslov <vladbu@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In order to remove dependency on rtnl lock and allow concurrent tcf_proto
modification, extend tcf_proto with reference counter. Implement helper
get/put functions for tcf proto and use them to modify cls API to always
take reference to tcf_proto while using it. Only release reference to
parent chain after releasing last reference to tp.
Signed-off-by: Vlad Buslov <vladbu@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Extend tcf_chain with new filter_chain_lock mutex. Always lock the chain
when accessing filter_chain list, instead of relying on rtnl lock.
Dereference filter_chain with tcf_chain_dereference() lockdep macro to
verify that all users of chain_list have the lock taken.
Rearrange tp insert/remove code in tc_new_tfilter/tc_del_tfilter to execute
all necessary code while holding chain lock in order to prevent
invalidation of chain_info structure by potential concurrent change. This
also serializes calls to tcf_chain0_head_change(), which allows head change
callbacks to rely on filter_chain_lock for synchronization instead of rtnl
mutex.
Signed-off-by: Vlad Buslov <vladbu@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently, tcf_block doesn't use any synchronization mechanisms to protect
critical sections that manage lifetime of its chains. block->chain_list and
multiple variables in tcf_chain that control its lifetime assume external
synchronization provided by global rtnl lock. Converting chain reference
counting to atomic reference counters is not possible because cls API uses
multiple counters and flags to control chain lifetime, so all of them must
be synchronized in chain get/put code.
Use single per-block lock to protect block data and manage lifetime of all
chains on the block. Always take block->lock when accessing chain_list.
Chain get and put modify chain lifetime-management data and parent block's
chain_list, so take the lock in these functions. Verify block->lock state
with assertions in functions that expect to be called with the lock taken
and are called from multiple places. Take block->lock when accessing
filter_chain_list.
In order to allow parallel update of rules on single block, move all calls
to classifiers outside of critical sections protected by new block->lock.
Rearrange chain get and put functions code to only access protected chain
data while holding block lock:
- Rearrange code to only access chain reference counter and chain action
reference counter while holding block lock.
- Extract code that requires block->lock from tcf_chain_destroy() into
standalone tcf_chain_destroy() function that is called by
__tcf_chain_put() in same critical section that changes chain reference
counters.
Signed-off-by: Vlad Buslov <vladbu@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
There are now several places where qdisc_tree_reduce_backlog() is called
with a negative number of packets (to signal an increase in number of
packets in the queue). Rather than rely on overflow behaviour, change the
function signature to use signed integers to communicate this usage to
people reading the code.
Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently drivers can register to receive TC block bind/unbind callbacks
by implementing the setup_tc ndo in any of their given netdevs. However,
drivers may also be interested in binds to higher level devices (e.g.
tunnel drivers) to potentially offload filters applied to them.
Introduce indirect block devs which allows drivers to register callbacks
for block binds on other devices. The callback is triggered when the
device is bound to a block, allowing the driver to register for rules
applied to that block using already available functions.
Freeing an indirect block callback will trigger an unbind event (if
necessary) to direct the driver to remove any offloaded rules and unreg
any block rule callbacks. It is the responsibility of the implementing
driver to clean any registered indirect block callbacks before exiting,
if the block it still active at such a time.
Allow registering an indirect block dev callback for a device that is
already bound to a block. In this case (if it is an ingress block),
register and also trigger the callback meaning that any already installed
rules can be replayed to the calling driver.
Signed-off-by: John Hurley <john.hurley@netronome.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Qdisc graft operation of offload-capable qdiscs performs a few
extra steps which are identical among all the qdiscs. Add
a helper to share this code.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: John Hurley <john.hurley@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Qdisc dump operation of offload-capable qdiscs performs a few
extra steps which are identical among all the qdiscs. Add
a helper to share this code.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: John Hurley <john.hurley@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Daniel Borkmann says:
====================
pull-request: bpf-next 2018-09-25
The following pull-request contains BPF updates for your *net-next* tree.
The main changes are:
1) Allow for RX stack hardening by implementing the kernel's flow
dissector in BPF. Idea was originally presented at netconf 2017 [0].
Quote from merge commit:
[...] Because of the rigorous checks of the BPF verifier, this
provides significant security guarantees. In particular, the BPF
flow dissector cannot get inside of an infinite loop, as with
CVE-2013-4348, because BPF programs are guaranteed to terminate.
It cannot read outside of packet bounds, because all memory accesses
are checked. Also, with BPF the administrator can decide which
protocols to support, reducing potential attack surface. Rarely
encountered protocols can be excluded from dissection and the
program can be updated without kernel recompile or reboot if a
bug is discovered. [...]
Also, a sample flow dissector has been implemented in BPF as part
of this work, from Petar and Willem.
[0] http://vger.kernel.org/netconf2017_files/rx_hardening_and_udp_gso.pdf
2) Add support for bpftool to list currently active attachment
points of BPF networking programs providing a quick overview
similar to bpftool's perf subcommand, from Yonghong.
3) Fix a verifier pruning instability bug where a union member
from the register state was not cleared properly leading to
branches not being pruned despite them being valid candidates,
from Alexei.
4) Various smaller fast-path optimizations in XDP's map redirect
code, from Jesper.
5) Enable to recognize BPF_MAP_TYPE_REUSEPORT_SOCKARRAY maps
in bpftool, from Roman.
6) Remove a duplicate check in libbpf that probes for function
storage, from Taeung.
7) Fix an issue in test_progs by avoid checking for errno since
on success its value should not be checked, from Mauricio.
8) Fix unused variable warning in bpf_getsockopt() helper when
CONFIG_INET is not configured, from Anders.
9) Fix a compilation failure in the BPF sample code's use of
bpf_flow_keys, from Prashant.
10) Minor cleanups in BPF code, from Yue and Zhong.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Implement get/put function for blocks that only take/release the reference
and perform deallocation. These functions are intended to be used by
unlocked rules update path to always hold reference to block while working
with it. They use on new fine-grained locking mechanisms introduced in
previous patches in this set, instead of relying on global protection
provided by rtnl lock.
Extract code that is common with tcf_block_detach_ext() into common
function __tcf_block_put().
Extend tcf_block with rcu to allow safe deallocation when it is accessed
concurrently.
Signed-off-by: Vlad Buslov <vladbu@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
As a preparation for removing rtnl lock dependency from rules update path,
change tcf block reference counter type to refcount_t to allow modification
by concurrent users.
In block put function perform decrement and check reference counter once to
accommodate concurrent modification by unlocked users. After this change
tcf_chain_put at the end of block put function is called with
block->refcnt==0 and will deallocate block after the last chain is
released, so there is no need to manually deallocate block in this case.
However, if block reference counter reached 0 and there are no chains to
release, block must still be deallocated manually.
Signed-off-by: Vlad Buslov <vladbu@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Implement function to take reference to Qdisc that relies on rcu read lock
instead of rtnl mutex. Function only takes reference to Qdisc if reference
counter isn't zero. Intended to be used by unlocked cls API.
Signed-off-by: Vlad Buslov <vladbu@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently, Qdisc API functions assume that users have rtnl lock taken. To
implement rtnl unlocked classifiers update interface, Qdisc API must be
extended with functions that do not require rtnl lock.
Extend Qdisc structure with rcu. Implement special version of put function
qdisc_put_unlocked() that is called without rtnl lock taken. This function
only takes rtnl lock if Qdisc reference counter reached zero and is
intended to be used as optimization.
Signed-off-by: Vlad Buslov <vladbu@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Current implementation of qdisc_destroy() decrements Qdisc reference
counter and only actually destroy Qdisc if reference counter value reached
zero. Rename qdisc_destroy() to qdisc_put() in order for it to better
describe the way in which this function currently implemented and used.
Extract code that deallocates Qdisc into new private qdisc_destroy()
function. It is intended to be shared between regular qdisc_put() and its
unlocked version that is introduced in next patch in this series.
Signed-off-by: Vlad Buslov <vladbu@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Adds a hook for programs of type BPF_PROG_TYPE_FLOW_DISSECTOR and
attach type BPF_FLOW_DISSECTOR that is executed in the flow dissector
path. The BPF program is per-network namespace.
Signed-off-by: Petar Penkov <ppenkov@google.com>
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Change flower in_hw_count type to fixed-size u32 and dump it as
TCA_FLOWER_IN_HW_COUNT. This change is necessary to properly test shared
blocks and re-offload functionality.
Signed-off-by: Vlad Buslov <vladbu@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Instead, adjust __qdisc_enqueue_tail() such that HTB can use it
instead.
The only other caller of __qdisc_enqueue_tail() is
qdisc_enqueue_tail() so we can move the backlog and return value
handling (which HTB doesn't need/want) to the latter.
Signed-off-by: David S. Miller <davem@davemloft.net>
This is similar TC_ACT_REDIRECT, but with a slightly different
semantic:
- on ingress the mirred skbs are passed to the target device
network stack without any additional check not scrubbing.
- the rcu-protected stats provided via the tcf_result struct
are updated on error conditions.
This new tcfa_action value is not exposed to the user-space
and can be used only internally by clsact.
v1 -> v2: do not touch TC_ACT_REDIRECT code path, introduce
a new action type instead
v2 -> v3:
- rename the new action value TC_ACT_REINJECT, update the
helper accordingly
- take care of uncloned reinjected packets in XDP generic
hook
v3 -> v4:
- renamed again the new action value (JiriP)
v4 -> v5:
- fix build error with !NET_CLS_ACT (kbuild bot)
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Each lockless action currently does its own RCU locking in ->act().
This allows using plain RCU accessor, even if the context
is really RCU BH.
This change drops the per action RCU lock, replace the accessors
with the _bh variant, cleans up a bit the surrounding code and
documents the RCU status in the relevant header.
No functional nor performance change is intended.
The goal of this patch is clarifying that the RCU critical section
used by the tc actions extends up to the classifier's caller.
v1 -> v2:
- preserve rcu lock in act_bpf: it's needed by eBPF helpers,
as pointed out by Daniel
v3 -> v4:
- fixed some typos in the commit message (JiriP)
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In case a chain is empty and not explicitly created by a user,
such chain should not exist. The only exception is if there is
an action "goto chain" pointing to it. In that case, don't show the
chain in the dump. Track the chain references held by actions and
use them to find out if a chain should or should not be shown
in chain dump.
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Allow user to set a template for newly created chains. Template lock
down the chain for particular classifier type/options combinations.
The classifier needs to support templates, otherwise kernel would
reply with error.
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Allow user to create, destroy, get and dump chain objects. Do that by
extending rtnl commands by the chain-specific ones. User will now be
able to explicitly create or destroy chains (so far this was done only
automatically according the filter/act needs and refcounting). Also, the
user will receive notification about any chain creation or destuction.
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently, chain 0 is implicitly created during block creation. However
that does not align with chain object exposure, creation and destruction
api introduced later on. So make the chain 0 behave the same way as any
other chain and only create it when it is needed. Since chain 0 is
somehow special as the qdiscs need to hold pointer to the first chain
tp, this requires to move the chain head change callback infra to the
block structure.
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add the reoffload tcf_proto_op in flower to generate an offload message
for each filter in the given tcf_proto. Call the specified callback with
this new offload message. The function only returns an error if the
callback rejects adding a 'hardware only' rule.
A filter contains a flag to indicate if it is in hardware or not. To
ensure the reoffload function properly maintains this flag, keep a
reference counter for the number of instances of the filter that are in
hardware. Only update the flag when this counter changes from or to 0. Add
a generic helper function to implement this behaviour.
Signed-off-by: John Hurley <john.hurley@netronome.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Create a new tcf_proto_op called 'reoffload' that generates a new offload
message for each node in a tcf_proto. Pointers to the tcf_proto and
whether the offload request is to add or delete the node are included.
Also included is a callback function to send the offload message to and
the option of priv data to go with the cb.
Signed-off-by: John Hurley <john.hurley@netronome.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
AFAICT struct gnet_stats_queue.qlen is not used in Qdiscs.
It may, however, be useful for offloads to report HW queue
length there. Add that value to the result of qdisc_qlen_sum().
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The struct Qdisc has a lot of holes, especially after commit
a53851e2c3 ("net: sched: explicit locking in gso_cpu fallback"),
which as a side effect, moved the fields just after 'busylock'
on a new cacheline.
Since both 'padded' and 'refcnt' are not updated frequently, and
there is a hole before 'gso_skb', we can move such fields there,
saving a cacheline without any performance side effect.
Before this commit:
pahole -C Qdisc net/sche/sch_generic.o
# ...
/* size: 384, cachelines: 6, members: 25 */
/* sum members: 236, holes: 3, sum holes: 92 */
/* padding: 56 */
After this commit:
pahole -C Qdisc net/sche/sch_generic.o
# ...
/* size: 320, cachelines: 5, members: 25 */
/* sum members: 236, holes: 2, sum holes: 28 */
/* padding: 56 */
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
So that we can use lockdep on it.
The newly introduced sequence lock has the same scope of busylock,
so it shares the same lockdep annotation, but it's only used for
NOLOCK qdiscs.
With this changeset we acquire such lock in the control path around
flushing operation (qdisc reset), to allow more NOLOCK qdisc perf
improvement in the next patch.
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently NOLOCK qdiscs pay a measurable overhead to atomically
manipulate the __QDISC_STATE_RUNNING. Such bit is flipped twice per
packet in the uncontended scenario with packet rate below the
line rate: on packed dequeue and on the next, failing dequeue attempt.
This changeset moves the bit manipulation into the qdisc_run_{begin,end}
helpers, so that the bit is now flipped only once per packet, with
measurable performance improvement in the uncontended scenario.
This also allows simplifying the qdisc teardown code path - since
qdisc_is_running() is now effective for each qdisc type - and avoid a
possible race between qdisc_run() and dev_deactivate_many(), as now
the some_qdisc_is_busy() can properly detect NOLOCK qdiscs being busy
dequeuing packets.
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Minor conflicts in drivers/net/ethernet/mellanox/mlx5/core/en_rep.c,
we had some overlapping changes:
1) In 'net' MLX5E_PARAMS_LOG_{SQ,RQ}_SIZE -->
MLX5E_REP_PARAMS_LOG_{SQ,RQ}_SIZE
2) In 'net-next' params->log_rq_size is renamed to be
params->log_rq_mtu_frames.
3) In 'net-next' params->hard_mtu is added.
Signed-off-by: David S. Miller <davem@davemloft.net>
After the qdisc lock was dropped in pfifo_fast we allow multiple
enqueue threads and dequeue threads to run in parallel. On the
enqueue side the skb bit ooo_okay is used to ensure all related
skbs are enqueued in-order. On the dequeue side though there is
no similar logic. What we observe is with fewer queues than CPUs
it is possible to re-order packets when two instances of
__qdisc_run() are running in parallel. Each thread will dequeue
a skb and then whichever thread calls the ndo op first will
be sent on the wire. This doesn't typically happen because
qdisc_run() is usually triggered by the same core that did the
enqueue. However, drivers will trigger __netif_schedule()
when queues are transitioning from stopped to awake using the
netif_tx_wake_* APIs. When this happens netif_schedule() calls
qdisc_run() on the same CPU that did the netif_tx_wake_* which
is usually done in the interrupt completion context. This CPU
is selected with the irq affinity which is unrelated to the
enqueue operations.
To resolve this we add a RUNNING bit to the qdisc to ensure
only a single dequeue per qdisc is running. Enqueue and dequeue
operations can still run in parallel and also on multi queue
NICs we can still have a dequeue in-flight per qdisc, which
is typically per CPU.
Fixes: c5ad119fb6 ("net: sched: pfifo_fast use skb_array")
Reported-by: Jakob Unterwurzacher <jakob.unterwurzacher@theobroma-systems.com>
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fun set of conflict resolutions here...
For the mac80211 stuff, these were fortunately just parallel
adds. Trivially resolved.
In drivers/net/phy/phy.c we had a bug fix in 'net' that moved the
function phy_disable_interrupts() earlier in the file, whilst in
'net-next' the phy_error() call from this function was removed.
In net/ipv4/xfrm4_policy.c, David Ahern's changes to remove the
'rt_table_id' member of rtable collided with a bug fix in 'net' that
added a new struct member "rt_mtu_locked" which needs to be copied
over here.
The mlxsw driver conflict consisted of net-next separating
the span code and definitions into separate files, whilst
a 'net' bug fix made some changes to that moved code.
The mlx5 infiniband conflict resolution was quite non-trivial,
the RDMA tree's merge commit was used as a guide here, and
here are their notes:
====================
Due to bug fixes found by the syzkaller bot and taken into the for-rc
branch after development for the 4.17 merge window had already started
being taken into the for-next branch, there were fairly non-trivial
merge issues that would need to be resolved between the for-rc branch
and the for-next branch. This merge resolves those conflicts and
provides a unified base upon which ongoing development for 4.17 can
be based.
Conflicts:
drivers/infiniband/hw/mlx5/main.c - Commit 42cea83f95
(IB/mlx5: Fix cleanup order on unload) added to for-rc and
commit b5ca15ad7e (IB/mlx5: Add proper representors support)
add as part of the devel cycle both needed to modify the
init/de-init functions used by mlx5. To support the new
representors, the new functions added by the cleanup patch
needed to be made non-static, and the init/de-init list
added by the representors patch needed to be modified to
match the init/de-init list changes made by the cleanup
patch.
Updates:
drivers/infiniband/hw/mlx5/mlx5_ib.h - Update function
prototypes added by representors patch to reflect new function
names as changed by cleanup patch
drivers/infiniband/hw/mlx5/ib_rep.c - Update init/de-init
stage list to match new order from cleanup patch
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Stanislav Fomichev