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: NVlad Buslov <vladbu@mellanox.com>
Acked-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid 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: NVlad Buslov <vladbu@mellanox.com>
Acked-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid 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: NVlad Buslov <vladbu@mellanox.com>
Acked-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid 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: NVlad Buslov <vladbu@mellanox.com>
Acked-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid 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: NVlad Buslov <vladbu@mellanox.com>
Acked-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid 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: NVlad Buslov <vladbu@mellanox.com>
Acked-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid 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: NVlad Buslov <vladbu@mellanox.com>
Acked-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid 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: NVlad Buslov <vladbu@mellanox.com>
Acked-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid 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: NToke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: NDavid 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: NJohn Hurley <john.hurley@netronome.com>
Signed-off-by: NJakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: NDavid 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: NJakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: NJohn Hurley <john.hurley@netronome.com>
Signed-off-by: NDavid 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: NJakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: NJohn Hurley <john.hurley@netronome.com>
Signed-off-by: NDavid 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: NVlad Buslov <vladbu@mellanox.com>
Acked-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid 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: NVlad Buslov <vladbu@mellanox.com>
Acked-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid 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: NVlad Buslov <vladbu@mellanox.com>
Acked-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid 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: NVlad Buslov <vladbu@mellanox.com>
Acked-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid 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: NVlad Buslov <vladbu@mellanox.com>
Acked-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid 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: NPetar Penkov <ppenkov@google.com>
Signed-off-by: NWillem de Bruijn <willemb@google.com>
Signed-off-by: NAlexei 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: NVlad Buslov <vladbu@mellanox.com>
Acked-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

By hand copies of SKB list handlers do not belong in individual packet
schedulers.
Signed-off-by: NDavid 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: NDavid 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: NPaolo Abeni <pabeni@redhat.com>
Acked-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid 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: NPaolo Abeni <pabeni@redhat.com>
Acked-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid 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: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid 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: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid 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: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid 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: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid 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: NJohn Hurley <john.hurley@netronome.com>
Signed-off-by: NJakub Kicinski <jakub.kicinski@netronome.com>
Acked-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid 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: NJohn Hurley <john.hurley@netronome.com>
Signed-off-by: NJakub Kicinski <jakub.kicinski@netronome.com>
Acked-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid 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: NJakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The struct Qdisc has a lot of holes, especially after commit
a53851e2 ("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: NPaolo Abeni <pabeni@redhat.com>
Acked-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid 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: NPaolo Abeni <pabeni@redhat.com>
Signed-off-by: NDavid 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: NPaolo Abeni <pabeni@redhat.com>
Signed-off-by: NDavid 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: c5ad119f ("net: sched: pfifo_fast use skb_array")
Reported-by: NJakob Unterwurzacher <jakob.unterwurzacher@theobroma-systems.com>
Signed-off-by: NJohn Fastabend <john.fastabend@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

When we exceed current packets limit and we have more than one
segment in the list returned by skb_gso_segment(), netem drops
only the first one, skipping the rest, hence kmemleak reports:

unreferenced object 0xffff880b5d23b600 (size 1024):
  comm "softirq", pid 0, jiffies 4384527763 (age 2770.629s)
  hex dump (first 32 bytes):
    00 80 23 5d 0b 88 ff ff 00 00 00 00 00 00 00 00  ..#]............
    00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
  backtrace:
    [<00000000d8a19b9d>] __alloc_skb+0xc9/0x520
    [<000000001709b32f>] skb_segment+0x8c8/0x3710
    [<00000000c7b9bb88>] tcp_gso_segment+0x331/0x1830
    [<00000000c921cba1>] inet_gso_segment+0x476/0x1370
    [<000000008b762dd4>] skb_mac_gso_segment+0x1f9/0x510
    [<000000002182660a>] __skb_gso_segment+0x1dd/0x620
    [<00000000412651b9>] netem_enqueue+0x1536/0x2590 [sch_netem]
    [<0000000005d3b2a9>] __dev_queue_xmit+0x1167/0x2120
    [<00000000fc5f7327>] ip_finish_output2+0x998/0xf00
    [<00000000d309e9d3>] ip_output+0x1aa/0x2c0
    [<000000007ecbd3a4>] tcp_transmit_skb+0x18db/0x3670
    [<0000000042d2a45f>] tcp_write_xmit+0x4d4/0x58c0
    [<0000000056a44199>] tcp_tasklet_func+0x3d9/0x540
    [<0000000013d06d02>] tasklet_action+0x1ca/0x250
    [<00000000fcde0b8b>] __do_softirq+0x1b4/0x5a3
    [<00000000e7ed027c>] irq_exit+0x1e2/0x210

Fix it by adding the rest of the segments, if any, to skb 'to_free'
list. Add new __qdisc_drop_all() and qdisc_drop_all() functions
because they can be useful in the future if we need to drop segmented
GSO packets in other places.

Fixes: 6071bd1a ("netem: Segment GSO packets on enqueue")
Signed-off-by: NAlexey Kodanev <alexey.kodanev@oracle.com>
Acked-by: NNeil Horman <nhorman@tuxdriver.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Fix trivial spelling mistake "greater then" -> "greater than".
Signed-off-by: NGal Pressman <galp@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Introduce a new qdisc ops ->change_tx_queue_len() so that
each qdisc could decide how to implement this if it wants.
Previously we simply read dev->tx_queue_len, after pfifo_fast
switches to skb array, we need this API to resize the skb array
when we change dev->tx_queue_len.

To avoid handling race conditions with TX BH, we need to
deactivate all TX queues before change the value and bring them
back after we are done, this also makes implementation easier.

Cc: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: NCong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Propagate extack to cls->destroy callbacks when called from
non-error paths.  On error paths pass NULL to avoid overwriting
the failure message.
Signed-off-by: NJakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: NSimon Horman <simon.horman@netronome.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch adds extack support for classifier delete callback api. This
prepares to handle extack support inside each specific classifier
implementation.

Cc: David Ahern <dsahern@gmail.com>
Signed-off-by: NAlexander Aring <aring@mojatatu.com>
Acked-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch adds extack support for classifier change callback api. This
prepares to handle extack support inside each specific classifier
implementation.

Cc: David Ahern <dsahern@gmail.com>
Signed-off-by: NAlexander Aring <aring@mojatatu.com>
Acked-by: NJiri Pirko <jiri@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>