| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
Custom ACL-plugin heap was useful in early stages,
but it interferes with other infra optimizations.
Remove it and use global heap.
Change-Id: I2300548f1d1798dec27bc5a2a41cf37f9fcda95d
Type: improvement
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
|
|
Type: style
and add some indent offs.
Change-Id: I31cf3ab9ff9b64d2cd1f2034dcedd4a9c453efb4
Signed-off-by: Neale Ranns <nranns@cisco.com>
|
|
implement per-acl-number counters in the stats segment.
They are created during the ACL creation,
the counters are incremented in the dataplane using
the new inline function with the extra parameter being
the packet size. Counting in shared segment adds
a noticeable overhead, so add also an API to
turn the counters on.
Type: feature
Change-Id: I8af7b0c31a3d986b68089eb52452aed45df66c7b
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
|
|
Long time ago, the linear array of rules in the ACL structure was not
a vector. Now it is, so get rid of the extraneous "count" member.
Do so in a manner that would ease potential the MP-safe manipulation of
ACL rules in the future.
Change-Id: Ib9c0731e4f21723c9ec4d7f00c3e5ead8e1e97bd
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
|
|
memory sizes [VPP-1502]
In a couple of places vec_add1()-style was repeatedly called in a loop for
smallish vectors where the number of additions was known in advance.
With a test with large number of ACEs these numbers contribute to heap
fragmentation noticeably.
Minimize the number of allocations by preallocating the known size and
then resetting the length accordingly, and then calling vec_add1()
Also unify the parsing of the memory-related startup config parameters.
Change-Id: If8fba344eb1dee8f865ffe7b396ca3b6bd9dc1d0
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
|
|
for plumbing
This makes ACL plugin use the new feature arcs, which slightly increases performance.
Since for ethertype whitelisting we were using the L2 classifier, to retain
the functionality, make a simple node doing that, and plug it into non-ip
L2 feature arc whenever needed.
Change-Id: I3add377a6c790117dd3fd056e5615cb4c4438cf4
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
|
|
This commit adds explicit signaling from a non-owning thread to the owning thread
to restart the session timer as necessary.
Consequently, we now can sweep the session lists at their respective timeouts,
rather than sweeping all the lists at the pace of the shortest timeout value,
just taking care to wake up if the session requeue to a different list results
in needing to wake up earlier.
Change-Id: Ifc8c500f6988748f4cd3dc184dd7824321aaaaca
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
|
|
Change-Id: I085615fde1f966490f30ed5d32017b8b088cfd59
Signed-off-by: Paul Vinciguerra <pvinci@vinciconsulting.com>
|
|
applied ACLs
The partition_split() did not increment the refcount when using a mask type index,
thus subsequent modifications potentially resulted in double frees and in the best case
immediate crash, in the worst case delayed crash in another place.
Introduce the lock_mask_type_index() and call it, move the mask type index
related functions closer to the top of the file.
Make the assignment of the new mask type indices
for the tuplemerge case to use the assign_mask_type_index().
Keep some debugs in case we need to investigate this further at some point.
Change-Id: Iae370f5cd92e1fe1442480db34656a8a3442dbc0
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
(cherry picked from commit 1edc406da3d4f6e63de2f278360b5753f55c00df)
|
|
Slightly refactored from the initial implementation of the TupleMerge [1]
algorithm by Valerio Bruschi (valerio.bruschi@telecom-paristech.fr)
[1] James Daly, Eric Torng "TupleMerge: Building Online Packet Classifiers
by Omitting Bits", In Proc. IEEE ICCCN 2017, pp. 1-10
Also add startup parameters to turn on/off the algorithm ("use tuple merge 1/0"),
and a startup parameter to be able to tweak the split threshold
("tuple merge split threshold N"), the default value of the split threshold
is 39 as per paper, but some more tuning might be necessary to find the best
value.
This change, alongside with the optimizations which avoid extra lookups,
significantly reduces the slowdown on the ClassBench generated ACLs, which
are supposed to resemble realistic ACLs seen in use in the field.
Change-Id: I9713e4673970e9a62d4d9e9718365293375fab7b
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
|
|
- instantiate the per-use mask type entry for a given hash ACE
this prepares to adding tuplemerge where the applied ACE may
have a different mask type due to relaxing of the tuples
- store the vector of the colliding rules for linear lookups
rather than traversing the linked list.
- store the lowest rule index for a given mask type inside
the structure. This allows to skip looking up at the later
mask types if we already matched an entry that is in front
of the very first entry in the new candidate mask type,
thus saving a worthless hash table lookup.
- use a vector of mask type indices rather than bitmap,
in the sorted order (by construction) of ascending
lowest rule index - this allows to terminate the lookups
early.
- adapt the debug cli outputs accordingly to show the data
- propagate the is_ip6 into the inner calls
Change-Id: I7a67b271e66785c6eab738b632b432d5886a0a8a
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
|
|
in more than one C file
Including the exports.h from multiple .c files belonging to a single plugin results in an error.
Rework the approach to require the table of function pointers to be filled in by
the initialization function.
Since the inline functions are compiled in the "caller" context,
there is no knowledge about the acl_main structure used by the ACL
plugin. To help with that, the signature of inline functions is slightly
different, taking the p_acl_main pointer as the first parameter.
That pointer is filled into the .p_acl_main field of the method
table during the initialization - since the calling of non-inline variants
would have required filling the method table, this should give
minimal headaches during the use and switch between the two methods.
Change-Id: Icb70695efa23579c46c716944838766cebc8573e
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
|
|
Add a new kv_16_8 field into 5tuple union, rename
the existing kv into kv_40_8 for clarity, and
add the compile-time alignment constraints.
Change-Id: I9bfca91f34850a5c89cba590fbfe9b865e63ef94
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
|
|
- implement a 1us purgatory for the session structures
by adding a special connection list, where all connections
about to be deleted go.
- add per-list-head timeouts updated upon the list enqueue/dequeue
for connection idle management
- add a "unused" session list with list ID#0, which should
never be used unless there is a logic error. Use this ID
to initialize the sessions.
- improve the maintainability of the session linked list
structures by using symbolic bogus index name instead of ~0
- change the ordering of session creations - first reverse, then
local. To minimize the potential for two workers competing for
the same session in the corner case of the two packets
on different workers creating the same logical session
- reduce the maximum session count to keep the memory usage the same
- add extra log/debug/trace to session cleaning logic
- be more aggressive with cleaning up sessions - wind up the
interrupts from the workers to themselves if there is more
work to do
Change-Id: I3aa1c91a925a08e83793467cb15bda178c21e426
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
|
|
Object sizes must evenly divide alignment requests, or vice
versa. Otherwise, only the first object will be aligned as
requested.
Three choices: add CLIB_CACHE_LINE_ALIGN_MARK(align_me) at
the end of structures, manually pad to an even divisor or multiple of
the alignment request, or use plain vectors/pools.
static assert for enforcement.
Change-Id: I41aa6ff1a58267301d32aaf4b9cd24678ac1c147
Signed-off-by: Dave Barach <dbarach@cisco.com>
|
|
using the inline functions
The acl_main struct, which is defined in the acl_plugin, is not visible when
the ACL plugin inline code is being compiled within the context of other plugins.
Fix that by using the global pointer variable, which exists in both the ACL plugin
context and is set in the context of the external plugins using ACL plugin.
Change-Id: Iaa74dd8cf36ff5442a06a25c5c968722116bddf8
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
(cherry picked from commit 1286a15a6e60f80b0e1b349f876de8fa38c71368)
|
|
(re-)applied
There were several discussions in which users would expect the sessions to be deleted
if the new policy after the change does not permit them.
There is no right or wrong answer to this question - it is a policy decision.
This patch implements an idea to approach this. It uses a per-interface-per-direction counter to designate
a "policy epoch" - a period of unchanging rulesets. The moment one removes or adds an ACL applied to
an interface, this counter increments.
The newly created connections inherit the current policy epoch in a given direction.
Likewise, this counter increments if anyone updates an ACL applied to an interface.
There is also a new (so far hidden) CLI "set acl-plugin reclassify-sessions [0|1]"
(with default being 0) which allows to enable the checking of the existing sessions
against the current policy epoch in a given direction.
The session is not verified unless there is traffic hitting that session
*in the direction of the policy creation* - if the epoch has changed,
the session is deleted and within the same processing cycle is evaluated
against the ACL rule base and recreated - thus, it should allow traffic-driven
session state refresh without affecting the connectivity for the existing sessions.
If the packet is coming in the direction opposite to which the session was initially
created, the state adjustment is never done, because doing so generically
is not really possible without diving too deep into the special cases,
which may or may not work.
Change-Id: I9e90426492d4bd474b5e89ea8dfb75a7c9de2646
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
|
|
- autosize the ACL plugin heap size based on the number of workers
- for manual heap size setting, use the proper types (uword),
and proper format/unformat functions (unformat_memory_size)
Change-Id: I7c46134e949862a0abc9087d7232402fc5a95ad8
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
|
|
- allow to optionally specify the specific MACIP ACL index:
'show acl-plugin macip acl [index N]'
- after showing the MACIP ACL, show the sw_if_index of
interface(s) where it is applied.
Also, add some executions of this debug commands
to the MACIP test case for easy verification.
Change-Id: I56cf8272abc20b1b2581b60d528d27a70d186b18
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
|
|
The functions which get called by other plugins need to set the acl plugin heap,
such that the other plugins do not have to think about it.
Change-Id: I673073f17116ffe444c163bf3dff40821d0c2686
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
|
|
This reverts commit 378ac0533e5ac8c3121d8f66ba61a8548e55282f.
Change-Id: If34b1c964453adb0e4c44e3eab4f6e306bd9c9e9
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
other plugins
This code implements the functionality required for other plugins wishing
to perform ACL lookups in the contexts of their choice, rather than only
in the context of the interface in/out.
The lookups are the stateless ACLs - there is no concept of "direction"
within the context, hence no concept of "connection" either.
The plugins need to include the
The file acl_lookup_context.md has more info.
Change-Id: I91ba97428cc92b24d1517e808dc2fd8e56ea2f8d
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
|
|
- Show interface on which given MACIP ACL is applied
- index is added for show acl-plugin macip acl:
ex) show acl-plugin macip acl [index N]
Change-Id: I3e888c8e3267060fe157dfc1bbe3e65371bd858a
Signed-off-by: Steve Shin <jonshin@cisco.com>
|
|
Currently, ACL plugin largely does not care about the
ethertypes other than 0x0800 (IPv4) and 0x86dd (IPv6),
the only exception being 0x0806 (ARP), which is
dealt with by the MACIP ACLs.
The other ethertypes in L2 mode are just let through.
This adds a new API message acl_interface_set_etype_whitelist,
which allows to flip the mode of a given interface
into "ethertype whitelist mode": the caller of this message
must supply the two lists (inbound and outbound) of the ethertypes
that are to be permitted, the rest of the ethertypes are
dropped.
The whitelisting for a given interface and direction takes
effect only when a policy ACL is also applied.
This operates on the same classifier node as the one used for
dispatching the policy ACL, thus, if one wishes for most of the
reasonable IPv4 deployments to continue to operate within
the whitelist mode, they must permit ARP ethertype (0x0806)
The empty list for a given direction resets the processing
to allow the unknown ethertypes. So, if one wants to just
permit the IPv4 and IPv6 and nothing else, one can add
their ethertypes to the whitelist.
Add the "show acl-plugin interface" corresponding outputs
about the whitelists, vat command, and unittests.
Change-Id: I4659978c801f36d554b6615e56e424b77876662c
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
|
|
This is the second patch, using the new functionality from the change 10002
in order to implement the egress filtering on the MACIP ACLs.
This adds an action "2" which means "add also egress filtering rules for this
MACIP ACL.
The reason for having the two choices is that the egress filtering really takes
care of a fairly corner case scenario, and I am not convinced that
always adding the performance cost of the egress lookup check is worth it.
Also, of course, not breaking the existing implementations is a nice plus,
too.
Change-Id: I3d7883ed45b1cdf98d7303771bcc75951dff38f0
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
|
|
conn cleaner threads interactions
This replaces some of the early-stage commented-out printf()s with
an elog-based debug collector.
It is aimed to be "better than nothing" initial implementation to be available
in the field. It will be refined/updated based on use. This initial code
is focused on the main/worker threads interactions, hence uses just
the worker tracks.
This code adds a developer debug CLI "set acl-plugin session table event-trace 1",
which allows to gather the events pertaining to connection cleaning.
The CLI is deliberately not part of the online help, as the express
declaration that the semantics/trace levels, etc. are subject to change
without notice.
Change-Id: I3536309f737b73e50639cd5780822dcde667fc2c
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
|
|
Change-Id: I92b351895c7efb26533c05512b91ead8ddbfb9c8
Signed-off-by: Pavel Kotucek <pkotucek@cisco.com>
|
|
This adds the ability to tweak the memory allocation parameters of the ACL plugin
from the startup config. It may be useful in the cases involving higher limit
of the connections than the default 1M, or the high number of cores.
Change-Id: I2b6fb3f61126ff3ee998424b762b6aefe8fb1b8e
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
|
|
Change-Id: I30a3df53bc5fe5ab991a657918eb502bd2913440
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
|
|
In multithread setup the main thread may send packets,
which may pass through the node with permit+reflect action.
This creates the connection in lists for thread0,
however in multithread there are no interupt handlers there.
Ensure we are not spending too much time spinning in a
tight cycle by suspending the main cleaner thread
until the current iteration of interrupts is processed.
Change-Id: Idb7346737757ee9a67b5d3e549bc9ad9aab22e89
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
(cherry picked from commit c1ff53f25d04ec1cc31844abd38014e91e398b5f)
|
|
(VPP-910)
The further prolonged testing from testbed that reported VPP-910
has uncovered a couple of deeper issues with optimization from
7384, and the usage of subscripts rather than vec_elt_at_index()
allowed to hide a couple of further errors in the code.
Also, the current acl-plugin behavior of using the global
heap for its dynamic data is problematic - it makes
the troubleshooting much harder by potentially spreading
the problem around.
Based on this experience, this commits makes a few changes to fix
the issues seen, also improving the serviceability of the acl-plugin
code for the future:
- Use separate mheaps for any ACL-related control plane
operations and separate for the hash lookup datastructures,
to compartmentalize any memory-related issues for the ACL plugin.
- Ensure vec_elt_at_index() usage throughout the hash_lookup.c file.
- Use vectors rather than raw memory for storing the "ordinary" ACL rules.
- Rework the optimization from 7384 to use a separate tail pointer
rather than overloading the "prev" field.
- Make get_session_ptr() more conservative and adjust is_valid_session_ptr
accordingly
Change-Id: Ifda85193f361de5ed3782a4acd39622bd33c5830
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
(cherry picked from commit bd9c5ffe39e9ce61db95d74d150e07d738f24da1)
|
|
(VPP-912)
Fix several threading-related issues uncovered by the CSIT scale/performance test:
- make the per-interface add/del counters per-thread
- preallocate the per-worker session pools rather than
attempting to resize them within the datapath
- move the bihash initialization to the moment of ACL
being applied rather than later during the connection creation
- adjust the connection cleaning logic to not require
the signaling from workers to main thread
- make the connection lists check in the main thread robust against workers
updating the list heads at the same time
- add more information to "show acl-plugin sessions" to aid in debugging
Change-Id: If82ef715e4993614df11db5e9afa7fa6b522d9bc
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
(cherry picked from commit 8e4222fc7e23a478b021930ade3cb7d20938e398)
|
|
Add a bihash-based ACL lookup mechanism and make it a new default.
This changes the time required to lookup a 5-tuple match
from O(total_N_entries) to O(total_N_mask_types), where
"mask type" is an overall mask on the 5-tuple required
to represent an ACE.
For testing/comparison there is a temporary debug CLI
"set acl-plugin use-hash-acl-matching {0|1}", which,
when set to 0, makes the plugin use the "old" linear lookup,
and when set to 1, makes it use the hash-based lookup.
Based on the discussions on vpp-dev mailing list,
prevent assigning the ACL index to an interface,
when the ACL with that index is not defined,
also prevent deleting an ACL if that ACL is applied.
Also, for the easier debugging of the state, there are
new debug CLI commands to see the ACL plugin state at
several layers:
"show acl-plugin acl [index N]" - show a high-level
ACL representation, used for the linear lookup and
as a base for building the hashtable-based lookup.
Also shows if a given ACL is applied somewhere.
"show acl-plugin interface [sw_if_index N]" - show
which interfaces have which ACL(s) applied.
"show acl-plugin tables" - a lower-level debug command
used to see the state of all of the related data structures
at once. There are specifiers possible, which make
for a more focused and maybe augmented output:
"show acl-plugin tables acl [index N]"
show the "bitmask-ready" representations of the ACLs,
we well as the mask types and their associated indices.
"show acl-plutin tables mask"
show the derived mask types and their indices only.
"show acl-plugin tables applied [sw_if_index N]"
show the table of all of the ACEs applied for a given
sw_if_index or all interfaces.
"show acl-plugin tables hash [verbose N]"
show the 48x8 bihash used for the ACL lookup.
Change-Id: I89fff051424cb44bcb189e3cee04c1b8f76efc28
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
|
|
A bihash-per-interface is convenient, but turns out tricky difficult from
the maintenance standpoint with the large number of interfaces.
This patch makes the sessions reside in a single hash table for all the interfaces,
adding the lower 16 bit of sw_if_index as part of the key into the previously
unused space.
There is a tradeoff, that a session with an identical 5-tuple and the same
sw_if_index modulo 65536 will match on either of the interfaces.
The probability of that is deemed sufficiently small to not worry about it.
In case it still happens before the heat death of the universe,
there is a clib_warning and the colliding packet will be dropped,
at which point we will need to bump the hash key size by another u64,
but rather not pay the cost of doing that right now.
Change-Id: I2747839cfcceda73e597cbcafbe1e377fb8f1889
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
|
|
code use it
This fixes the undesirable pause in the dump commands in case there is nothing to dump.
Change-Id: I0554556c9e442038aa2a1ed8c88234f21f7fe9b9
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
|
|
Add the logic to be able to use stateful ACLs in a multithreaded setup.
Change-Id: I3b0cfa6ca4ea8f46f61648611c3e97b00c3376b6
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
|
|
Change-Id: I3d64d5ced38a68f3fa208be00c49d20c4e6d4d0e
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
|
|
(VPP-682)
This fixes the previously-implicit "drop all non-first fragments" behavior
to be more in line with security rules: a non-first fragment is treated
for the purposes of matching the ACL as a packet with the port
match succeeding. This allows to change the behavior to permit
the fragmented packets for the default "permit specific rules"
ruleset, but also gives the flexibility to block the non-initial
fragments by inserting into the begining a bogus rule
which would deny the L4 traffic.
Also, add a knob which allows to potentially turn this behavior off
in case of a dire need (and revert to dropping all non-initial fragments),
via a debug CLI.
Change-Id: I546b372b65ff2157d9c68b1d32f9e644f1dd71b4
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
(cherry picked from commit 9fc0c26c6b28fd6c8b8142ea52f52eafa7e8c7ac)
|
|
- use the counters in a private struct rather than node error counters
- ensure the timer for the non-idle connections is restarted
- fix the deletion of conn at the current tail the list
Change-Id: I632f63574d2ced95fb75c5e7fb588c78fb3cce1c
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
(cherry picked from commit 097051a3bd1f63a177c0728f15375afd84a68918)
|
|
L3 path support, L2+L3 unified processing node, skip IPv6 EH support.
Change-Id: Iac37a466ba1c035e5c2997b03c0743bfec5c9a08
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
|
|
via l2output_main.next_nodes
Before this commit, several output features that happen to be the
last in the list of features to be executed, send the packets directly
to <interfaceName>-output. To do this, they use l2_output_dispatch,
which builds a list of sw_if_index to next index mappings.
When interfaces are deleted and the new interfaces are created,
these mappings become stale, and cause the packets being sent to wrong
interface output nodes.
This patch (thanks John Lo for the brilliant idea!) adds a feature node "output",
whose sole purpose is dispatching the packets to the correct interface output
nodes. To do that, it uses the l2output_main.next_nodes, which is already
taken care of for the case of the sw_if_index reuse, so this makes the dependent
features all work correctly.
Since this changes the packet path, for the features that were always the last ones
it has triggered a side problem of the output feat_next_node_index not being properly
initalized. These two users are l2-output-classify node and the output nodes belonging
to the acl-plugin.
For the first one the less invasive fix is just to initialize that field.
For the acl-plugin nodes, rewrite the affected part of the code to use
feat_bitmap_get_next_node_index since this is essentially what the conditional
in l2_output_dispatch does, and fix the compiler warnings generated.
This fix was first made in stable/1701 under commit e7dcee4027854b0ad076101471afdfff67eb9011.
Change-Id: I32e876ab1e1d498cf0854c19c6318dcf59a93805
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
|
|
This patch replaces requirement for vlib_plugin_register function
in the plugin so file and introduces new macro:
VLIB_PLUGIN_REGISTER () = {
.version = "version string",
.version_required = "requred version",
.default_disabled = 1,
.early_init = "early_init_function_name",
};
Plugin will nor be loaded if .default_disabled is set to 1
unless explicitely enabled in startup.conf.
If .verstion_required is set, plugin will not be loaded if there
is version mismatch between plugin and vpp. This can be bypassed
by setting "skip-version-check" for specific plugin.
If .early-init string is present, plugin loader will try to resolve
this specific symbol in the plugin namespace and make a function call.
Following startup.conf configuration is added:
plugins {
path /path/to/plugin/directory
plugin ila_plugin.so { enable skip-version-check }
plugin acl_plugin.so { disable }
}
Change-Id: I706c691dd34d94ffe9e02b59831af8859a95f061
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
|
Change-Id: I1c3b87e886603678368428ae56a6bd3327cbc90d
Signed-off-by: Damjan Marion <damarion@cisco.com>
|
Andrew Yourtchenko
Neale Ranns
Paul Vinciguerra
Dave Barach
Damjan Marion
Steve Shin
Pavel Kotucek