The commit reverted an earlier change which was fixed a bug that caused
black-holes to remote destinations with multiple paths, that could occur
during convergence. Overall, the previous code is more correct.
Avneesh Sachdev [Fri, 17 Aug 2012 15:19:50 +0000 (08:19 -0700)]
lib/table: add route_table_get_next() and iterator
* lib/table.[ch]
- Add a function (route_table_get_next()) to get the route_node in
a tree that succeeds a given prefix in iteration order.
This allows one to reliably walk nodes in a tree while allowing
modifications, and is useful for achieving scale and
performance. Other approaches are also possible -- the main plus
point of this one is that it does not require any state about
the walk to be maintained in the table data structures.
- Add an iterator for walking the nodes in a tree. This introduces
a new structure (route_table_iter_t) and the following main
functions.
The iterator normally uses node pointers and the existing
route_next() function to walk nodes efficiently. When an
iteration is 'paused' with route_table_iter_pause(), it stores
the last prefix processed. The next call to
route_table_iter_next() transparently invokes
route_table_get_next() with the prefix to resume iteration.
* bgpd/bgp_table.[ch]
Add wrappers for the new table features described above.
* tests/table_test.c
Add tests for the new table code.
Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
Avneesh Sachdev [Fri, 17 Aug 2012 15:19:49 +0000 (08:19 -0700)]
bgpd: make bgp_table a wrapper around table library
Make the BGP table code a thin wrapper around the table implementation
in libzebra.
* bgpd/bgp_table.[ch]
- Use the ROUTE_NODE_FIELDS macro to embed the fields of a
route_node in the bgp_node structure.
- Add a route_table field to the bgp_table structure.
Initialize the route_table with a delegate, such that the nodes
in the table are bgp_node structures.
- Add inline wrappers that call route_table functions underneath,
and accept/return the correct BGP types.
* bgpd/bgp_route.c
Change some code to use inline wrappers instead of accessing
fields of nodes/tables directly. The latter does not always work
because the types of some fields need to be translated now.
Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
Avneesh Sachdev [Fri, 17 Aug 2012 15:19:48 +0000 (08:19 -0700)]
lib: prepare table code for reuse by bgp_table
* lib/table.[ch]
- Add a macro (ROUTE_NODE_FIELDS) that expands to all the fields
of a route_node structure.
- Add the route_table_delegate_t structure, a function vector
which allows clients to customize the behavior of one or more
tables.
The delegate currently contains the 'create_node' and
'destroy_node' functions, and hence enables a table to use an
alternative node structure. The alternative node is expected to
embed the fields of a route_node using ROUTE_NODE_FIELDS.
- Add route_table_init_with_delegate() to create a new table with
a given delegate.
- Make route_table_init() a thin wrapper around
route_table_init_with_delegate(). The delegate it passes in
simply creates/destroys route_node structures as before.
- Add a user data pointer (info) to the route_table
structure. This can be used by a client to keep per-table state.
Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
The patch in bug 330 did two things. It add a return value
whether ospf_nexthop_calculation() failed or not and also
moved the return stmt for 16.1.1 para 5 so now SPF
will fallback to the intervening router when no back links are found
by 16.1.1 para 5. This is wrong and can potentially create black holes
or routing loops according to Dave Katz and Acee Lindem at ospf@ietf.org
Even if the current code could be proved to be harmless in all cases,
it adds substantial extra processing and memory allocations.
Signed-off-by: Joakim Tjernlund <Joakim.Tjernlund@transmode.se> Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
OSPF really needs to specify interface in its routes. Otherwise
ospf may change the wrong route.
Signed-off-by: Joakim Tjernlund <Joakim.Tjernlund@transmode.se>
[fixed up some whitespace errors, split patch in two] Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
Actually implement the IPV4_IFINDEX nexthop type that has been drifting
around as a definition forever (without any warning about it being a
placeholder).
Signed-off-by: Joakim Tjernlund <Joakim.Tjernlund@transmode.se> Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
ospfd: Optimize and improve SPF nexthop calculation
Maintain router LSA positions in OSPF interface.
Find the OSPF interface in nexthop_calculation using
the position in the router LSA. This is possible because
the only time nexthop_calculation needs to look up interfaces
is when dealing with its own Router LSA.
This has the following advantages:
- Multiple PtP interfaces with the same IP address between two routers.
- Use Unnumbered PtP on just one end of the link.
- Faster OI lookup for the OSPF interface and only
done once for PtoP links.
*ospf_interface.h: (struct ospf_interface) Add storage for
storing router LSA position.
*ospf_interface.c: (ospf_if_lookup_by_lsa_pos)
lookup OSPF I/F in an area using LSA position.
*ospf_lsa.c: (router_lsa_link_set) record Router LSA position.
*ospf_spf.c: (ospf_spf_next) Count and pass along lsa position.
(ospf_nexthop_calculation) Add lsa position argument.
call ospf_if_lookup_by_lsa_pos() for OSFP interface handle.
Clean up and remove all calls ospf_if_is_configured() the
rest. Adjust a few debug logs.
Signed-off-by: Joakim Tjernlund <Joakim.Tjernlund@transmode.se> Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
David Lamparter [Mon, 23 Jul 2012 16:17:57 +0000 (18:17 +0200)]
ospfd: avoid exhausting memory with OSPF vertices (BZ#476)
This was found in scale testing at OSR; ospfd is adding the same link
over and over again to the SPF tree. This fix prevents the resulting
memory corruption from happening and adds a debug message to track
occurence of this issue and/or confirm a proper fix.
(This version was improved by Scott Feldman over the earlier RFC.)
* ospfd/ospf_spf.c: (ospf_spf_add_parent) loop over existing vertices
and refuse to add duplicates.
Tested-by: Martin Winter <mwinter@opensourcerouting.org> Signed-off-by: Scott Feldman <sfeldma@cumulusnetworks.com> Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
David Lamparter [Thu, 19 Jul 2012 14:11:50 +0000 (16:11 +0200)]
lib: make IPv6 prefix parser slightly more strict
This makes it possible to have both "show babel route A.B.C.D/M"
and "show babel route X:X::X:X/M" commands at the same time without
the parser complaining about ambiguity.
* lib/command.c: only accept STATE_DOT after : was seen.
Reported-by: Juliusz Chroboczek <jch@pps.jussieu.fr> Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
David Lamparter [Mon, 1 Feb 2010 15:41:26 +0000 (16:41 +0100)]
zebra: lingering IP address after deletion (BZ#486)
zebra address bookkeeping is a mess. this is just a workaround to have
IPv4 address deletion somewhat working on Linux.
the if_unset_prefix call is synchronous, when it returns success the
address deletion completed successfully. this is either signaled by a
netlink ACK or by an OK return value from ioctl().
This version is wrapped by #ifdef HAVE_NETLINK so we don't touch the
BSDs for now.
* zebra/interface.c: On Linux, update zebra internal state after
deleting an address.
Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
Joachim Nilsson [Wed, 30 May 2012 06:15:48 +0000 (08:15 +0200)]
configure: Add --with-pkg-extra-version=VER for packagers/distributions.
This change adds a --with-pkg-extra-version option to ./configure to allow
packagers and distributions to fine tune the version displayed to the end
user to assist in support. It is also very useful when reporting bugs on
the official Quagga mailing lists.
There are two ways of utilsing this functionality:
a) ./configure --with-pkg-extra-version=-wmo1
b) EXTRAVERSION=-wmo1 ./configure
The latter is a common way for many distributions to add extra version
strings to signify their own patch level to a given package.
Also, minor whitespace fix in configure summary.
Signed-off-by: Joachim Nilsson <joachim.nilsson@westermo.se> Signed-off-by: Joachim Nilsson <troglobit@gmail.com> Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
Vincent Bernat [Tue, 22 May 2012 11:34:24 +0000 (13:34 +0200)]
snmp: fix OSPFV3-MIB implementation
Use the real MIB from RFC 5643. Fix used ASN1 types. Indexes are not
exported any more (they are "no access"). Fix some endian
issues. Also, ID are just integers, not IPv4 addresses.
Vincent Bernat [Thu, 31 May 2012 11:30:28 +0000 (13:30 +0200)]
snmp: let handlers accept OID from a lesser prefix
Most table handlers do not expect to be given an OID whose prefix is
outside what they can handle. This is not a problem with the SMUX
implementation since it always correct the OID such that the prefix
matches. However, this is not the case for the AgentX
implementation. A new function, smux_header_table() is used to do this
normalization.
Vincent Bernat [Fri, 25 May 2012 09:17:01 +0000 (11:17 +0200)]
agentx: handle SNMP traps
smux_trap() signature has been changed to provide appropriate level
information to send SNMPv2 notifications. This includes the addition
of the enterprise OID to use (from which is derived the SNMP trap OID)
and the MIB registry to locate the appropriate function for variable
bindings provided by the trap.
The SMUX implementation has been updated but ignore the provided
enterprise OID. Instead, it still uses the SMUX peer OID to keep
compatibility with previous versions of Quagga. The SMUX
implementation also ignores the provided MIB registry since it uses
smux_get() function to grab the appropriate values. This is not
possible with the AgentX implementation since there is no such
function provided by NetSNMP.
Vincent Bernat [Thu, 24 May 2012 19:22:01 +0000 (21:22 +0200)]
smux: remove `tick` argument from smux_trap()
smux_trap() contains an argument whose use appears to be to set
sysUpTime.0/timestamp field in SNMP trap. However, this value is not
used in smux_trap(). Moreover, it is expected that this field is the
value of sysUpTime.0 when the trap was sent and not any other time
related to the trap. To avoid any confusion, we remove this field from
the signature of the function.
Vincent Bernat [Thu, 24 May 2012 07:44:43 +0000 (09:44 +0200)]
agentx: add AgentX support to Quagga.
--enable-snmp will enable AgentX support in Quagga. SMUX is still here
and can be enabled with --enable-snmp=smux. AgentX support can be
enabled with "agentx" in configuration file. As for SMUX, this command
is not understood by vtysh. It can be disabled with "no agentx",
though there is no real use of this since this command cannot be used
with vtysh.
If "agentx" and "no agentx" command were added to vtysh, it would not
be possible to disable agentx support after enabling it because
NetSNMP does not expose the appropriate methods for this.
The internals of AgentX are hidden by NetSNMP. Therefore, we don't
have a file descriptor to add to the threading system. We do not have
the timers to set either. Therefore, the event loop is modified to
make use of snmp_select_info() from NetSNMP.
Vincent Bernat [Tue, 22 May 2012 22:52:46 +0000 (00:52 +0200)]
smux: isolate SMUX implementation from SNMP implementation
lib/snmp.c gets OID related helper functions that can be used with
another SNMP interface. smux.h is cleaned of SMUX specific bits to
only expose functions that may be used by an alternative
implementation. We also do not redefine functions already present in
NetSNMP. Just use the appropriate headers.
Vincent Bernat [Tue, 22 May 2012 20:29:17 +0000 (22:29 +0200)]
build: use net-snmp-config to configure NetSNMP
The correct method to link to NetSNMP is to use net-snmp-config (which
is like pkg-config). Explicit link to libcrypto is also dropped
(NetSNMP libs are linked to libcrypto, no need to link Quagga to
it). Moreover, @SNMP_INCLUDES@ is dropped because useless. Due to a
bug in configure.ac, it was properly populated.
Vincent Bernat [Tue, 22 May 2012 12:32:22 +0000 (14:32 +0200)]
build: allow configure and build in a separate directory
Some .h files in lib/ are autogenerated. The search path should
include the build directory and the source directory. They usually
match but sometimes, they may be different. For example:
Paul Jakma [Thu, 14 Jun 2012 09:42:39 +0000 (10:42 +0100)]
tests: BGP unit tests to set BGP_OPT_NO_LISTEN so they can run
BGP tests had been broken by auto-creation of listen socket. This allows
them to run at least, though at least 1 test seems to have other breakage,
aspath_test.
Paul Jakma [Thu, 14 Jun 2012 09:40:26 +0000 (10:40 +0100)]
bgpd: Add "no listen" socket option for the BGP master configuration
* bgpd.h: add a BGP_OPT_NO_LISTEN option for the master BGP configuration,
to prevent any listen socket being created automatically. Allows code
to be used outside of BGP daemon settings.
* bgpd.c: (bgp_get) honour above the flag, suppress auto-creation of listen
socket on first BGP instance if set.
(bgp_option_set) whitelist BGP_OPT_NO_LISTEN
Paul Jakma [Thu, 14 Jun 2012 09:37:40 +0000 (10:37 +0100)]
lib: Add back sockunion_str2su
* sockunion.h: Add back sockunion_str2su, its removal breaks things needlessly
(e.g. our own unit tests).
* sockunion.c: (sockunion_str2su) implement on top of str2sockunion.
This avoids memory heap fragmentation and imposses less load on the
system memory allocator.
* thread.h: FUNCNAME_LEN defined to 64 (ISO C99 says max 63)
Signed-off-by: Jorge Boncompte [DTI2] <jorge@dti2.net>
[changed FUNCNAME_LEN to a less arbitrary value] Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
If there were no aggregates configured this functions were allocating
and freeing a struct bgp_node for every call, and it's called for every
prefix received.
* bgp_route.c: Bail out early if the there are no aggregates configured.
Change from bgp_node_get() to bgp_node_lookup() that does not allocate
a new struct bgp_node if not found.
Signed-off-by: Jorge Boncompte [DTI2] <jorge@dti2.net> Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
* bgp_packet.c: (bgp_update_receive) for every update received we queue
an event just to cancel the holdtime timer, done in bgp_fsm_update().
Instead cancel the timer directly an avoid a scheduling pass.
This incidently fixes another problem found on a slow box, where thousands
of events threads were queued, and run, but never freed, because they are
moved to the unused list that grows without bounds.
Signed-off-by: Jorge Boncompte [DTI2] <jorge@dti2.net> Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
peer_sort() it's called so much as to be annoying. In the assumption
that the 'sort' of the peer doesn't change during an established session,
I have changed all calls to peer_sort() in the 'fast-path' to only check
the 'sort'. All the calls from the vty and such still recalculate the sort
and store it in the peer.
There's a lot of other calls to peer_sort() that could be changed but some
maube tricky, someone more knowledgeable may try to reduce them.
This hits peer_sort() from 5th out of the stadium^H^H list on a full
internet table loading profiling session.
Signed-off-by: Jorge Boncompte [DTI2] <jorge@dti2.net> Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
Just the first change pushes bgp_update_receive() from 6th to ~14th on a
full internet table load profiling session.
* bgp_debug.c: (bgp_update_receive) The attrstr initialization is expensive,
moved under the debug conditional where it is used and just initialize the
first char to NULL.
(bgp_update_default_send) Initialize attrstr needed for bgp_dump_attr().
Moved some buffers used for printing IP[4|6] addresses under the debug
conditionals that use them and reduced its size.
Signed-off-by: Jorge Boncompte [DTI2] <jorge@dti2.net> Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
This function scores 2nd, profiling a full internet table load. It's called
for every prefix received.
Instead of looping in the interface lists comparing addresses use a hash
to mantain them.
* bgpd.c: Init the own address hash.
* bgp_nexthop.c: Introduce methods to maintain an own address hash.
(bgp_connected_add) add addresses to the hash.
(bgp_connected_delete) delete addresses from the hash.
(bgp_nexthop_self) lookup addresses in the hash. Removed the unused afi_t
parameter.
* bgp_route.c: (bgp_update_main) Micro-optimization, rearranged condition to
not lookup the hash for bogus nexthops (0.0.0.0 or a class D/E address)
Signed-off-by: Jorge Boncompte [DTI2] <jorge@dti2.net> Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
bgpd: optimize aspath string representation and assegments handling
* bgp_aspath.h: Add str_len to struct aspath.
* bgp_aspath.c: Save the aspath string representation length and use it
instead of strlen().
(aspath_make_str_count) assign the string buffer directly for
consistency with the string length and change the return type to void.
(aspath_dup) use str_len and copy the string instead of calling
aspath_make_str_count().
(assegment_data_new) change from XCALLOC to XMALLOC. All users initialize
the memory before use.
(assegment_data_free) unused, removed.
(aspath_intern) check that there's always a ->str pointer.
(aspath_hash_alloc) reuse assegments and string representation instead of
copying them.
(aspath_parse) now aspath_hash_alloc does not dupes memory, free the
temporary structures only if the aspath it is in the hash.
(aspath_cmp_left) remove useless NULL initialization.
Signed-off-by: Jorge Boncompte [DTI2] <jorge@dti2.net> Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
ospfd was crashing some times on neighbour going down. The cause was that
ospf_nsm_event() was accessing already freed memory in ospf_nbr_delete()
call from ospf_nsm_event().
What happens is that since commit b5043aab (lib: fix incorrect thread
list...) now a thread can be on the event and ready lists but
thread_cancel_event() doesn't account for that.
* thread.c: (thread_cancel_event) loop on the ready list too to cancel
pending events.
Signed-off-by: Jorge Boncompte [DTI2] <jorge@dti2.net> Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
* command.c: (cmd_ipv6_match) Drop IPv6 address recognition
heuristics and solely rely on inet_pton, because strings
like "abcd" were mistaken for IPv6 addresses.
This affects e.g. the command "neighbour WORD peer-group",
which won't work with words consisting of up to 4 characters
between 'a' and 'f' and digits.
From: Roman Hoog Antink <rha@open.ch>
[full delete instead of #if 0] Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
Joachim Nilsson [Wed, 9 May 2012 11:38:36 +0000 (13:38 +0200)]
zebra: fix up compilation without rtadv/IPv6
let's ground the rtadv.h file if route advertisements are disabled. And
fix up the CLI for it, as well as move the "show ip mroute" to its
proper place.
* zebra/rtadv.h: #ifdef RTADV
* zebra/main.c: #ifdef RTADV
* zebra/zebra_vty.c: move "show ip mroute" out of #ifdef IPV6
From: Joachim Nilsson <troglobit@gmail.com>
[moved #ifdef RTADV to rtadv.h] Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
The SAFI_MULTICAST RIB needs to contain Connected routes so that the
nexthop lookup does not fail and so that multicast routing daemons do
not need to sidestep and look into SAFI_UNICAST to be aware of connected
subnets.
* zebra/connected.c: add & delete connected in SAFI_MULTICAST
Signed-off-by: G.Balaji <balajig81@gmail.com>
[merged add/delete patches] Signed-off-by: David Lamparter <equinox@diac24.net>
zebra: no need to clear memory of the netlink buffers
The memory is always written before sending. Clearing it before use
slows netlink_route_multipath() down considerably and it's pretty
noticeable in a full-internet table scenario loading.
* zebra/rt_netlink.c: subtract bulk buffer size from clear
Paul Jakma [Tue, 1 May 2012 15:24:35 +0000 (16:24 +0100)]
bgpd: Make socket init separate, so unit tests work again.
* Separate out BGP socket initialisation from bgp_get, and make it an
explicit function. Allows unit tests to work again and probably also
benefits dry-run.
* bgpd.c: (bgp_get) move socket init out...
(bgp_socket_init) to here
* bgp_main.c: and call it after dry-run.
* bgpd.h: (bgp_socket_init) add prototype
Paul Jakma [Tue, 1 May 2012 15:20:33 +0000 (16:20 +0100)]
tests: Fix some compile errors and warnings
* aspath_test.c: match changes in aspath_unintern. Fix printf size_t warning.
* bgp_capability_test.c: compile warnings.
* bgp_mp_attr_test.c: update for attr parser context struct
* ecommunity_test.c: ecommunity_free/ecommunity
* test-checksum.c: some unused vars and funcs without need of prototypes.
David Ward [Sun, 29 Apr 2012 20:47:08 +0000 (16:47 -0400)]
redhat: add systemd service files
* redhat/*.service: add systemd service file for each routing daemon
* redhat/quagga.sysconfig: add comment about watchquagga and systemd
* redhat/Makefile.am: add systemd service files to distribution
Signed-off-by: David Ward <david.ward@ll.mit.edu> Signed-off-by: David Lamparter <equinox@diac24.net>
David Ward [Sun, 29 Apr 2012 20:47:07 +0000 (16:47 -0400)]
redhat: update initscripts
Taken from Fedora packaging with additional fixes.
* redhat/*.init: make all initscripts LSB-compliant;
store the daemon configuration file location in $CONF_FILE,
and perform existence testing before starting the daemon
* redhat/babeld.init: add initscript for Babel routing engine
* redhat/quagga.sysconfig: add command-line options for babeld;
remove the daemon configuration file locations from $*_OPTS;
clarify directions for configuring watchquagga
* redhat/Makefile.am: add babeld.init to distribution
Signed-off-by: David Ward <david.ward@ll.mit.edu> Signed-off-by: David Lamparter <equinox@diac24.net>
David Lamparter [Sat, 28 Apr 2012 20:37:20 +0000 (22:37 +0200)]
bgpd: fix up multipath merge breakage
the BGP multipath code was developed against a pre-f6f434b Quagga where
bgp_attr_unintern and ecommunity_free took single-star pointers. They
now take double-star pointers.
* bgpd/bgp_mpath.c: fix up bgp_attr_unintern & ecommunity_free calls.
Signed-off-by: David Lamparter <equinox@diac24.net>
In upstream babel, we include different kernel_foo files depending on the
platform we compile for. In Quagga, this is not needed, and it avoids
tricky autotools issues. (All autotools issues are tricky.)
[David: without this commit, build breaks.]
From: Juliusz Chroboczek <jch@pps.jussieu.fr> Signed-off-by: David Lamparter <equinox@diac24.net>
David Lamparter [Fri, 20 Apr 2012 15:26:48 +0000 (17:26 +0200)]
lib: bump ZSERV_VERSION to 2
continually changing the zserv protocol without bumping up the version
number has made it impossible to talk to zebra without knowing the exact
version. in reality, increasing the version number more often guards
against inadvertedly running incompatible versions of a daemon and zebra
as well as aids external development.
* lib/zclient.h: #define ZSERV_VERSION 2
Signed-off-by: David Lamparter <equinox@diac24.net>
projects / mirror_frr.git / log