This patch fixes netfilter hook traversal when there are more than 1 hooks
returning NF_QUEUE verdict. When the first queue reinjects the packet,
'nf_reinject' starts traversing hooks with a proper hook_index. However,
if it again receives a NF_QUEUE verdict (by some other netfilter hook), it
queues the packet with a wrong hook_index. So, when the second queue
reinjects the packet, it re-executes hooks in between.
Fixes: 960632ece6 ("netfilter: convert hook list to an array")
Signed-off-by: Jagdish Motwani <jagdish.motwani@sophos.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Replace NF_HOOK() based invocation of the netfilter hooks with a private
copy of nf_hook_slow().
This copy has one difference: it can return the rx handler value expected
by the stack, i.e. RX_HANDLER_CONSUMED or RX_HANDLER_PASS.
This is needed by the next patch to invoke the ebtables
"broute" table via the standard netfilter hooks rather than the custom
"br_should_route_hook" indirection that is used now.
When the skb is to be "brouted", we must return RX_HANDLER_PASS from the
bridge rx input handler, but there is no way to indicate this via
NF_HOOK(), unless perhaps by some hack such as exposing bridge_cb in the
netfilter core or a percpu flag.
text data bss dec filename
3369 56 0 3425 net/bridge/br_input.o.before
3458 40 0 3498 net/bridge/br_input.o.after
This allows removal of the "br_should_route_hook" in the next patch.
Signed-off-by: Florian Westphal <fw@strlen.de>
Acked-by: David S. Miller <davem@davemloft.net>
Acked-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
This pointer is going to be removed soon, so use the existing helpers in
more places to avoid noise when the removal happens.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
We cannot make a direct call to nf_ip6_reroute() because that would result
in autoloading the 'ipv6' module because of symbol dependencies.
Therefore, define reroute indirection in nf_ipv6_ops where this really
belongs to.
For IPv4, we can indeed make a direct function call, which is faster,
given IPv4 is built-in in the networking code by default. Still,
CONFIG_INET=n and CONFIG_NETFILTER=y is possible, so define empty inline
stub for IPv4 in such case.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
This is only used by nf_queue.c and this function comes with no symbol
dependencies with IPv6, it just refers to structure layouts. Therefore,
we can replace it by a direct function call from where it belongs.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
no need to define hook points if the family isn't supported.
Because we need these hooks for either nftables, arp/ebtables
or the 'call-iptables' hack we have in the bridge layer add two
new dependencies, NETFILTER_FAMILY_{ARP,BRIDGE}, and have the
users select them.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
struct net contains:
struct nf_hook_entries __rcu *hooks[NFPROTO_NUMPROTO][NF_MAX_HOOKS];
which store the hook entry point locations for the various protocol
families and the hooks.
Using array results in compact c code when doing accesses, i.e.
x = rcu_dereference(net->nf.hooks[pf][hook]);
but its also wasting a lot of memory, as most families are
not used.
So split the array into those families that are used, which
are only 5 (instead of 13). In most cases, the 'pf' argument is
constant, i.e. gcc removes switch statement.
struct net before:
/* size: 5184, cachelines: 81, members: 46 */
after:
/* size: 4672, cachelines: 73, members: 46 */
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
since commit 960632ece6 ("netfilter: convert hook list to an array")
nfqueue no longer stores a pointer to the hook that caused the packet
to be queued. Therefore no extra synchronize_net() call is needed after
dropping the packets enqueued by the old rule blob.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
This converts the storage and layout of netfilter hook entries from a
linked list to an array. After this commit, hook entries will be
stored adjacent in memory. The next pointer is no longer required.
The ops pointers are stored at the end of the array as they are only
used in the register/unregister path and in the legacy br_netfilter code.
nf_unregister_net_hooks() is slower than needed as it just calls
nf_unregister_net_hook in a loop (i.e. at least n synchronize_net()
calls), this will be addressed in followup patch.
Test setup:
- ixgbe 10gbit
- netperf UDP_STREAM, 64 byte packets
- 5 hooks: (raw + mangle prerouting, mangle+filter input, inet filter):
empty mangle and raw prerouting, mangle and filter input hooks:
353.9
this patch:
364.2
Signed-off-by: Aaron Conole <aconole@bytheb.org>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
queued skbs might be using conntrack extensions that are being removed,
such as timeout. This happens for skbs that have a skb->nfct in
unconfirmed state (i.e., not in hash table yet).
This is destructive, but there are only two use cases:
- module removal (rare)
- netns cleanup (most likely no conntracks exist, and if they do,
they are removed anyway later on).
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
nf_unregister_net_hook(s) can avoid a second call to synchronize_net,
provided there is no nfqueue active in that net namespace (which is
the common case).
This also gets rid of the extra arg to nf_queue_nf_hook_drop(), normally
this gets called during netns cleanup so no packets should be queued.
For the rare case of base chain being unregistered or module removal
while nfqueue is in use the extra hiccup due to the packet drops isn't
a big deal.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
nf_iterate() has become rather simple, we can integrate this code into
nf_hook_slow() to reduce the amount of LOC in the core path.
However, we still need nf_iterate() around for nf_queue packet handling,
so move this function there where we only need it. I think it should be
possible to refactor nf_queue code to get rid of it definitely, but
given this is slow path anyway, let's have a look this later.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
This field is only useful for nf_queue, so store it in the
nf_queue_entry structure instead, away from the core path. Pass
hook_head to nf_hook_slow().
Since we always have a valid entry on the first iteration in
nf_iterate(), we can use 'do { ... } while (entry)' loop instead.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Patch c5136b15ea ("netfilter: bridge: add and use br_nf_hook_thresh")
introduced br_nf_hook_thresh().
Replace NF_HOOK_THRESH() by br_nf_hook_thresh from
br_nf_forward_finish(), so we have no more callers for this macro.
As a result, state->thresh and explicit thresh parameter in the hook
state structure is not required anymore. And we can get rid of
skip-hook-under-thresh loop in nf_iterate() in the core path that is
only used by br_netfilter to search for the filter hook.
Suggested-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
nf_queue handling is broken since e3b37f11e6 ("netfilter: replace
list_head with single linked list") for two reasons:
1) If the bypass flag is set on, there are no userspace listeners and
we still have more hook entries to iterate over, then jump to the
next hook. Otherwise accept the packet. On nf_reinject() path, the
okfn() needs to be invoked.
2) We should not re-enter the same hook on packet reinjection. If the
packet is accepted, we have to skip the current hook from where the
packet was enqueued, otherwise the packets gets enqueued over and
over again.
This restores the previous list_for_each_entry_continue() behaviour
happening from nf_iterate() that was dealing with these two cases.
This patch introduces a new nf_queue() wrapper function so this fix
becomes simpler.
Fixes: e3b37f11e6 ("netfilter: replace list_head with single linked list")
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
The netfilter hook list never uses the prev pointer, and so can be trimmed to
be a simple singly-linked list.
In addition to having a more light weight structure for hook traversal,
struct net becomes 5568 bytes (down from 6400) and struct net_device becomes
2176 bytes (down from 2240).
Signed-off-by: Aaron Conole <aconole@bytheb.org>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Florian Weber reported:
> Under full load (unshare() in loop -> OOM conditions) we can
> get kernel panic:
>
> BUG: unable to handle kernel NULL pointer dereference at 0000000000000008
> IP: [<ffffffff81476c85>] nfqnl_nf_hook_drop+0x35/0x70
> [..]
> task: ffff88012dfa3840 ti: ffff88012dffc000 task.ti: ffff88012dffc000
> RIP: 0010:[<ffffffff81476c85>] [<ffffffff81476c85>] nfqnl_nf_hook_drop+0x35/0x70
> RSP: 0000:ffff88012dfffd80 EFLAGS: 00010206
> RAX: 0000000000000008 RBX: ffffffff81add0c0 RCX: ffff88013fd80000
> [..]
> Call Trace:
> [<ffffffff81474d98>] nf_queue_nf_hook_drop+0x18/0x20
> [<ffffffff814738eb>] nf_unregister_net_hook+0xdb/0x150
> [<ffffffff8147398f>] netfilter_net_exit+0x2f/0x60
> [<ffffffff8141b088>] ops_exit_list.isra.4+0x38/0x60
> [<ffffffff8141b652>] setup_net+0xc2/0x120
> [<ffffffff8141bd09>] copy_net_ns+0x79/0x120
> [<ffffffff8106965b>] create_new_namespaces+0x11b/0x1e0
> [<ffffffff810698a7>] unshare_nsproxy_namespaces+0x57/0xa0
> [<ffffffff8104baa2>] SyS_unshare+0x1b2/0x340
> [<ffffffff81608276>] entry_SYSCALL_64_fastpath+0x1e/0xa8
> Code: 65 00 48 89 e5 41 56 41 55 41 54 53 83 e8 01 48 8b 97 70 12 00 00 48 98 49 89 f4 4c 8b 74 c2 18 4d 8d 6e 08 49 81 c6 88 00 00 00 <49> 8b 5d 00 48 85 db 74 1a 48 89 df 4c 89 e2 48 c7 c6 90 68 47
>
The simple fix for this requires a new pernet variable for struct
nf_queue that indicates when it is safe to use the dynamically
allocated nf_queue state.
As we need a variable anyway make nf_register_queue_handler and
nf_unregister_queue_handler pernet. This allows the existing logic of
when it is safe to use the state from the nfnetlink_queue module to be
reused with no changes except for making it per net.
The syncrhonize_rcu from nf_unregister_queue_handler is moved to a new
function nfnl_queue_net_exit_batch so that the worst case of having a
syncrhonize_rcu in the pernet exit path is not experienced in batch
mode.
Reported-by: Florian Westphal <fw@strlen.de>
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Acked-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
All verdict handlers make use of the nfnetlink .call_rcu callback
so rcu readlock is already held.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
We don't care if module is being unloaded anymore since hook unregister
handling will destroy queue entries using that hook.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
since commit 8405a8fff3 ("netfilter: nf_qeueue: Drop queue entries on
nf_unregister_hook") all pending queued entries are discarded.
So we can simply remove all of the owner handling -- when module is
removed it also needs to unregister all its hooks.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Usage of -prev seems buggy. While packet was out our hook cannot be
removed but we have no way to know if the previous one is still valid.
So better not use ->prev at all. Since NF_REPEAT just asks to invoke
same hook function again, just do so, and continue with nf_interate
if we get an ACCEPT verdict.
A side effect of this change is that if nf_reinject(NF_REPEAT) causes
another REPEAT we will now drop the skb instead of a kernel loop.
However, NF_REPEAT loops would be a bug so this should not happen anyway.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
The network namespace is needed when routing a packet.
Stop making nf_afinfo.reroute guess which network namespace
is the proper namespace to route the packet in.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
This is immediately motivated by the bridge code that chains functions that
call into netfilter. Without passing net into the okfns the bridge code would
need to guess about the best expression for the network namespace to process
packets in.
As net is frequently one of the first things computed in continuation functions
after netfilter has done it's job passing in the desired network namespace is in
many cases a code simplification.
To support this change the function dst_output_okfn is introduced to
simplify passing dst_output as an okfn. For the moment dst_output_okfn
just silently drops the struct net.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If someone sends packets from one of the netdevice ingress hooks to
the a userspace queue, and then userspace later accepts the packet,
the netfilter code can enter an infinite loop as the list head will
never be found.
Pass in the saved list_head to avoid this.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
right now we store this in the nf_bridge_info struct, accessible
via skb->nf_bridge. This patch prepares removal of this pointer from skb:
Instead of using skb->nf_bridge->x, we use helpers to obtain the in/out
device (or ifindexes).
Followup patches to netfilter will then allow nf_bridge_info to be
obtained by a call into the br_netfilter core, rather than keeping a
pointer to it in sk_buff.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
On the output paths in particular, we have to sometimes deal with two
socket contexts. First, and usually skb->sk, is the local socket that
generated the frame.
And second, is potentially the socket used to control a tunneling
socket, such as one the encapsulates using UDP.
We do not want to disassociate skb->sk when encapsulating in order
to fix this, because that would break socket memory accounting.
The most extreme case where this can cause huge problems is an
AF_PACKET socket transmitting over a vxlan device. We hit code
paths doing checks that assume they are dealing with an ipv4
socket, but are actually operating upon the AF_PACKET one.
Signed-off-by: David S. Miller <davem@davemloft.net>
It is currently always set to NULL, but nf_queue is adjusted to be
prepared for it being set to a real socket by taking and releasing a
reference to that socket when necessary.
Signed-off-by: David S. Miller <davem@davemloft.net>
That way we don't have to reinstantiate another nf_hook_state
on the stack of the nf_reinject() path.
Signed-off-by: David S. Miller <davem@davemloft.net>
Instead of passing a large number of arguments down into the nf_hook()
entry points, create a structure which carries this state down through
the hook processing layers.
This makes is so that if we want to change the types or signatures of
any of these pieces of state, there are less places that need to be
changed.
Signed-off-by: David S. Miller <davem@davemloft.net>
In 34666d4 ("netfilter: bridge: move br_netfilter out of the core"),
the bridge netfilter code has been modularized.
Use IS_ENABLED instead of ifdef to cover the module case.
Fixes: 34666d4 ("netfilter: bridge: move br_netfilter out of the core")
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
skb_gso_segment is expensive, so it would be nice if we could
avoid it in the future. However, userspace needs to be prepared
to receive larger-than-mtu-packets (which will also have incorrect
l3/l4 checksums), so we cannot simply remove it.
The plan is to add a per-queue feature flag that userspace can
set when binding the queue.
The problem is that in nf_queue, we only have a queue number,
not the queue context/configuration settings.
This patch should have no impact other than the skb_gso_segment
call now being in a function that has access to the queue config
data.
A new size attribute in nf_queue_entry is needed so
nfnetlink_queue can duplicate the entry of the gso skb
when segmenting the skb while also copying the route key.
The follow up patch adds switch to disable skb_gso_segment when
queue config says so.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
required by future patch that will need to duplicate the
nf_queue_entry, bumping refcounts of the copy.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Add copyright statements to all netfilter files which have had significant
changes done by myself in the past.
Some notes:
- nf_conntrack_ecache.c was incorrectly attributed to Rusty and Netfilter
Core Team when it got split out of nf_conntrack_core.c. The copyrights
even state a date which lies six years before it was written. It was
written in 2005 by Harald and myself.
- net/ipv{4,6}/netfilter.c, net/netfitler/nf_queue.c were missing copyright
statements. I've added the copyright statement from net/netfilter/core.c,
where this code originated
- for nf_conntrack_proto_tcp.c I've also added Jozsef, since I didn't want
it to give the wrong impression
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
We used to have several queueing backends, but nowadays only
nfnetlink_queue remains.
In light of this there doesn't seem to be a good reason to
support per-af registering -- just hook up nfnetlink_queue on module
load and remove it on unload.
This means that the userspace BIND/UNBIND_PF commands are now obsolete;
the kernel will ignore them.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Since 'list_for_each_continue_rcu' has already been replaced by
'list_for_each_entry_continue_rcu', pass 'list_head' to nf_queue() as a
parameter can not benefit us any more.
This patch will replace 'list_head' with 'nf_hook_ops' as the parameter of
nf_queue() and __nf_queue() to save code.
Signed-off-by: Michael Wang <wangyun@linux.vnet.ibm.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Since 'list_for_each_continue_rcu' has already been replaced by
'list_for_each_entry_continue_rcu', pass 'list_head' to nf_iterate() as a
parameter can not benefit us any more.
This patch will replace 'list_head' with 'nf_hook_ops' as the parameter of
nf_iterate() to save code.
Signed-off-by: Michael Wang <wangyun@linux.vnet.ibm.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
When trying to nf_queue GRO/GSO skbs, nf_queue uses skb_gso_segment
to split the skb.
However, if nf_queue is called via bridge netfilter, the mac header
won't be preserved -- packets will thus contain a bogus mac header.
Fix this by setting skb->data to the mac header when skb->nf_bridge
is set and restoring skb->data afterwards for all segments.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
commit a9b3cd7f32 (rcu: convert uses of rcu_assign_pointer(x, NULL) to
RCU_INIT_POINTER) did a lot of incorrect changes, since it did a
complete conversion of rcu_assign_pointer(x, y) to RCU_INIT_POINTER(x,
y).
We miss needed barriers, even on x86, when y is not NULL.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
CC: Stephen Hemminger <shemminger@vyatta.com>
CC: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When assigning a NULL value to an RCU protected pointer, no barrier
is needed. The rcu_assign_pointer, used to handle that but will soon
change to not handle the special case.
Convert all rcu_assign_pointer of NULL value.
//smpl
@@ expression P; @@
- rcu_assign_pointer(P, NULL)
+ RCU_INIT_POINTER(P, NULL)
// </smpl>
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If an skb is to be NF_QUEUE'd, but no program has opened the queue, the
packet is dropped.
This adds a v2 target revision of xt_NFQUEUE that allows packets to
continue through the ruleset instead.
Because the actual queueing happens outside of the target context, the
'bypass' flag has to be communicated back to the netfilter core.
Unfortunately the only choice to do this without adding a new function
argument is to use the target function return value (i.e. the verdict).
In the NF_QUEUE case, the upper 16bit already contain the queue number
to use. The previous patch reduced NF_VERDICT_MASK to 0xff, i.e.
we now have extra room for a new flag.
If a hook issued a NF_QUEUE verdict, then the netfilter core will
continue packet processing if the queueing hook
returns -ESRCH (== "this queue does not exist") and the new
NF_VERDICT_FLAG_QUEUE_BYPASS flag is set in the verdict value.
Note: If the queue exists, but userspace does not consume packets fast
enough, the skb will still be dropped.
Signed-off-by: Florian Westphal <fwestphal@astaro.com>
Signed-off-by: Patrick McHardy <kaber@trash.net>
NF_VERDICT_MASK is currently 0xffff. This is because the upper
16 bits are used to store errno (for NF_DROP) or the queue number
(NF_QUEUE verdict).
As there are up to 0xffff different queues available, there is no more
room to store additional flags.
At the moment there are only 6 different verdicts, i.e. we can reduce
NF_VERDICT_MASK to 0xff to allow storing additional flags in the 0xff00 space.
NF_VERDICT_BITS would then be reduced to 8, but because the value is
exported to userspace, this might cause breakage; e.g.:
e.g. 'queuenr = (1 << NF_VERDICT_BITS) | NF_QUEUE' would now break.
Thus, remove NF_VERDICT_BITS usage in the kernel and move the old value
to the 'userspace compat' section.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Patrick McHardy <kaber@trash.net>