Under high load instances with hundreds of thousands of prefixes this
could result in very unstable systems.
When maximum-prefix is set, but restart timer is not set then the session
flaps between Idle(Pfx) -> Established -> Idle(Pfx) states.
Signed-off-by: Donatas Abraitis <donatas.abraitis@gmail.com>
Our Address Sanitizer CI is finding this issue:
error 09-Oct-2019 19:28:33 r4: bgpd triggered an exception by AddressSanitizer
error 09-Oct-2019 19:28:33 ERROR: AddressSanitizer: stack-buffer-overflow on address 0x7ffdd425b060 at pc 0x00000068575f bp 0x7ffdd4258550 sp 0x7ffdd4258540
error 09-Oct-2019 19:28:33 READ of size 1 at 0x7ffdd425b060 thread T0
error 09-Oct-2019 19:28:33 #0 0x68575e in prefix_cmp lib/prefix.c:776
error 09-Oct-2019 19:28:33 #1 0x5889f5 in rfapiItBiIndexSearch bgpd/rfapi/rfapi_import.c:2230
error 09-Oct-2019 19:28:33 #2 0x5889f5 in rfapiBgpInfoFilteredImportVPN bgpd/rfapi/rfapi_import.c:3520
error 09-Oct-2019 19:28:33 #3 0x58b909 in rfapiProcessWithdraw bgpd/rfapi/rfapi_import.c:4071
error 09-Oct-2019 19:28:33 #4 0x4c459b in bgp_withdraw bgpd/bgp_route.c:3736
error 09-Oct-2019 19:28:33 #5 0x484122 in bgp_nlri_parse_vpn bgpd/bgp_mplsvpn.c:237
error 09-Oct-2019 19:28:33 #6 0x497f52 in bgp_nlri_parse bgpd/bgp_packet.c:315
error 09-Oct-2019 19:28:33 #7 0x49d06d in bgp_update_receive bgpd/bgp_packet.c:1598
error 09-Oct-2019 19:28:33 #8 0x49d06d in bgp_process_packet bgpd/bgp_packet.c:2274
error 09-Oct-2019 19:28:33 #9 0x6b9f54 in thread_call lib/thread.c:1531
error 09-Oct-2019 19:28:33 #10 0x657037 in frr_run lib/libfrr.c:1052
error 09-Oct-2019 19:28:33 #11 0x42d268 in main bgpd/bgp_main.c:486
error 09-Oct-2019 19:28:33 #12 0x7f806032482f in __libc_start_main (/lib/x86_64-linux-gnu/libc.so.6+0x2082f)
error 09-Oct-2019 19:28:33 #13 0x42bcc8 in _start (/usr/lib/frr/bgpd+0x42bcc8)
error 09-Oct-2019 19:28:33
error 09-Oct-2019 19:28:33 Address 0x7ffdd425b060 is located in stack of thread T0 at offset 240 in frame
error 09-Oct-2019 19:28:33 #0 0x483945 in bgp_nlri_parse_vpn bgpd/bgp_mplsvpn.c:103
error 09-Oct-2019 19:28:33
error 09-Oct-2019 19:28:33 This frame has 5 object(s):
error 09-Oct-2019 19:28:33 [32, 36) 'label'
error 09-Oct-2019 19:28:33 [96, 108) 'rd_as'
error 09-Oct-2019 19:28:33 [160, 172) 'rd_ip'
error 09-Oct-2019 19:28:33 [224, 240) 'prd' <== Memory access at offset 240 overflows this variable
error 09-Oct-2019 19:28:33 [288, 336) 'p'
error 09-Oct-2019 19:28:33 HINT: this may be a false positive if your program uses some custom stack unwind mechanism or swapcontext
error 09-Oct-2019 19:28:33 (longjmp and C++ exceptions *are* supported)
error 09-Oct-2019 19:28:33 SUMMARY: AddressSanitizer: stack-buffer-overflow lib/prefix.c:776 prefix_cmp
error 09-Oct-2019 19:28:33 Shadow bytes around the buggy address:
error 09-Oct-2019 19:28:33 0x10003a8435b0: 00 00 00 00 00 00 f1 f1 f1 f1 00 00 00 00 00 00
error 09-Oct-2019 19:28:33 0x10003a8435c0: 00 00 00 00 00 00 00 00 00 00 f3 f3 f3 f3 f3 f3
error 09-Oct-2019 19:28:33 0x10003a8435d0: f3 f3 00 00 00 00 00 00 00 00 00 00 00 00 00 00
error 09-Oct-2019 19:28:33 0x10003a8435e0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 f1 f1
error 09-Oct-2019 19:28:33 0x10003a8435f0: f1 f1 04 f4 f4 f4 f2 f2 f2 f2 00 04 f4 f4 f2 f2
error 09-Oct-2019 19:28:33 =>0x10003a843600: f2 f2 00 04 f4 f4 f2 f2 f2 f2 00 00[f4]f4 f2 f2
error 09-Oct-2019 19:28:33 0x10003a843610: f2 f2 00 00 00 00 00 00 f4 f4 f3 f3 f3 f3 00 00
error 09-Oct-2019 19:28:33 0x10003a843620: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
error 09-Oct-2019 19:28:33 0x10003a843630: 00 00 00 00 00 00 00 00 00 00 f1 f1 f1 f1 02 f4
error 09-Oct-2019 19:28:33 0x10003a843640: f4 f4 f2 f2 f2 f2 04 f4 f4 f4 f2 f2 f2 f2 00 00
error 09-Oct-2019 19:28:33 0x10003a843650: f4 f4 f2 f2 f2 f2 00 00 00 00 f2 f2 f2 f2 00 00
error 09-Oct-2019 19:28:33 Shadow byte legend (one shadow byte represents 8 application bytes):
error 09-Oct-2019 19:28:33 Addressable: 00
error 09-Oct-2019 19:28:33 Partially addressable: 01 02 03 04 05 06 07
error 09-Oct-2019 19:28:33 Heap left redzone: fa
error 09-Oct-2019 19:28:33 Heap right redzone: fb
error 09-Oct-2019 19:28:33 Freed heap region: fd
error 09-Oct-2019 19:28:33 Stack left redzone: f1
error 09-Oct-2019 19:28:33 Stack mid redzone: f2
error 09-Oct-2019 19:28:33 Stack right redzone: f3
error 09-Oct-2019 19:28:33 Stack partial redzone: f4
error 09-Oct-2019 19:28:33 Stack after return: f5
error 09-Oct-2019 19:28:33 Stack use after scope: f8
error 09-Oct-2019 19:28:33 Global redzone: f9
error 09-Oct-2019 19:28:33 Global init order: f6
error 09-Oct-2019 19:28:33 Poisoned by user: f7
error 09-Oct-2019 19:28:33 Container overflow: fc
error 09-Oct-2019 19:28:33 Array cookie: ac
error 09-Oct-2019 19:28:33 Intra object redzone: bb
error 09-Oct-2019 19:28:33 ASan internal: fe
error 09-Oct-2019 19:28:36 r3: Daemon bgpd not running
This is the result of this code pattern in rfapi/rfapi_import.c:
prefix_cmp((struct prefix *)&bpi_result->extra->vnc.import.rd,
(struct prefix *)prd))
Effectively prd or vnc.import.rd are `struct prefix_rd` which
are being typecast to a `struct prefix`. Not a big deal except commit
1315d74de9 modified the prefix_cmp
function to allow for a sorted prefix_cmp. In prefix_cmp
we were looking at the offset and shift. In the case
of vnc we were passing a prefix length of 64 which is the exact length of
the remaining data structure for struct prefix_rd. So we calculated
a offset of 8 and a shift of 0. The data structures for the prefix
portion happened to be equal to 64 bits of data. So we checked that
with the memcmp got a 0 and promptly read off the end of the data
structure for the numcmp. The fix is if shift is 0 that means thei
the memcmp has checked everything and there is nothing to do.
Please note: We will still crash if we set the prefixlen > then
~312 bits currently( ie if the prefixlen specifies a bit length
longer than the prefix length ). I do not think there is
anything to do here( nor am I sure how to correct this either )
as that we are going to have some severe problems when we muck
up the prefixlen.
Fixes: #5025
Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>
BGP code assumes that the extra data is zero'ed out. Ensure that we
are not leaving any situation that the data on the stack is actually all
0's when we pass it around as a pointer later.
Please note in issue #5025, Lou reported a different valgrind
issue, which is not the same issue:
==7313== Conditional jump or move depends on uninitialised value(s)
==7313== at 0x181F9F: subgroup_announce_check (bgp_route.c:1555)
==7313== by 0x1A112B: subgroup_announce_table (bgp_updgrp_adv.c:641)
==7313== by 0x1A1340: subgroup_announce_route (bgp_updgrp_adv.c:704)
==7313== by 0x1A13E3: subgroup_coalesce_timer (bgp_updgrp_adv.c:331)
==7313== by 0x4EBA615: thread_call (thread.c:1531)
==7313== by 0x4E8AC37: frr_run (libfrr.c:1052)
==7313== by 0x1429E0: main (bgp_main.c:486)
==7313==
==7313== Conditional jump or move depends on uninitialised value(s)
==7313== at 0x201C0E: rfapi_vty_out_vncinfo (rfapi_vty.c:429)
==7313== by 0x18D0D6: route_vty_out (bgp_route.c:7481)
==7313== by 0x18DD76: bgp_show_table (bgp_route.c:9365)
==7313== by 0x1930C4: bgp_show_table_rd (bgp_route.c:9471)
==7313== by 0x1932A3: bgp_show (bgp_route.c:9510)
==7313== by 0x193E68: show_ip_bgp_json (bgp_route.c:10284)
==7313== by 0x4E6D024: cmd_execute_command_real.isra.2 (command.c:1072)
==7313== by 0x4E6F51E: cmd_execute_command (command.c:1131)
==7313== by 0x4E6F686: cmd_execute (command.c:1285)
==7313== by 0x4EBF9C4: vty_command (vty.c:516)
==7313== by 0x4EBFB9F: vty_execute (vty.c:1285)
==7313== by 0x4EC250F: vtysh_read (vty.c:2119)
==7313==
that is causing the actual crash.
Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>
When using soft reconfiguration inbound we are storing packet
data on the side for replaying when necessary. The problem here
is that we are just grabbing the first bgp_path_info and using
that as the base. What happens when we have soft-reconfig turned
on with multiple bgp_path_info's for a path? This was introduced
in commit 8692c50652, yes back
in 2012! I would argue, though, that it was just broken
in a different way before this.
Choose the correct bgp_path_info that corresponds to the peer
we received the data from for rethinking.
Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>
ALL Daemons
-N <namespace> to allow for config file locating when running FRR
inside of a namespace
Impoved Testing across all daemons
BFD
VRF Support
Conversion to Northbound interface
BGP
Aggregate-address add route-map support
BMP Support
Improved JSON output for many commands
`show bgp afi safi summary failed` command
`clear bop *` clears all peers
Show FQDN for `show bgp ipv4 uni` commands
Display BestPath selection reason as part of show commands
EIGRP
Infrastructure changes to allow VRF's
SIGHUP signals the config reload
Conversion to Northbound interface
ISIS
BFD Support
Support for circuits with MTU > 8192
PBRD
fwmark support as part of match criteria
autocompletion of PBRMAPS
Improved Nexthop Support
PIMD
PIM-BSM receive support
Improved debugging support
Store ECMP paths that are not currently legal for use
Disallow igmp query from a non-connected source
Many new cli improvements and changes
VRRPD
Add Support for RFC 3768 and RFC 5798
Route-Maps
Add sequence numbers to access-lists
Add `match ip next-hop type blackhole`
Improved ability to notice dependency changes
SHARPD
`sharp watch [import|nexthop]` you can now specify a prefix instead
of assuming a /32
STATICD
Significantly Improved NHT
ZEBRA
Many dataplane improvements for routes, neighbor table and EVPN
NHT cli can now be specified per VRF and improved ability to control
NHT data being shown
Removed duplicate processing of routes
Improved debugablility
RMAC and VxLan support for the FPM
LIB
RCU support
Nexthop Group Improvements
`log-filter WORD` added
Building
openssl support
libcap should be used as part of build or significant slowdowns will
be experienced
Lua builds have been fixed
Improved Cross building
Snapcraft
Add Fabricd
Add Libyan
Update rtrlib and rpki
Signed-off-by: Martin Winter <mwinter@opensourcerouting.org>
When using link-local addresses we must provide scope-id to the
operating system so it knows where to send packets.
Spotted by Pavel Ivashchenko (@zays26).
Signed-off-by: Rafael Zalamena <rzalamena@opensourcerouting.org>
Don't be selective about what to observe, always observe all possible
aspects of the session that may change on run-time (i.e. bind address,
interface and VRF existence).
Signed-off-by: Rafael Zalamena <rzalamena@opensourcerouting.org>
If a session is no longer able to send/receive packets, it is very
likely it will be down in a few milliseconds so lets speed up the
process and correctly mark it as down.
Signed-off-by: Rafael Zalamena <rzalamena@opensourcerouting.org>
Otherwise the `thread_read` will keep waking us up to handle closing
sockets which are never unregistered.
Signed-off-by: Rafael Zalamena <rzalamena@opensourcerouting.org>
commit ee8a72f315
broke the usage of ZEBRA_ROUTE_ALL as a valid redistribution
command. This commit puts it back in. LDP uses ZEBRA_ROUTE_ALL
as an option to say it is interested in all REDISTRIBUTION events.
Fixes: #5072
Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>
The vrf_id in `zsend_interface_vrf_update()` is encoded as
a long via `stream_putl()`, we should decode it as such
as well.
Signed-off-by: Stephen Worley <sworley@cumulusnetworks.com>
RFC 5303 states:
If the system ID and Extended Local Circuit ID of the neighboring
system are known (in adjacency three-way state Initializing or
Up), the neighbor's system ID SHALL be reported in the Neighbor
System ID field, and the neighbor's Extended Local Circuit ID
SHALL be reported in the Neighbor Extended Local Circuit ID field.
There is nothing written about only setting the Extended circuit ID of the
adjacency only when we bring the three-way adjacency up.
In fact, we should always update it, to avoid the problem described in #4783.
Fixes: #4783
Signed-off-by: Christian Franke <chris@opensourcerouting.org>
Current autocompletion works only for simple "vrf NAME" case.
This commit expands it also for the following cases:
- "nexthop-vrf NAME" in staticd
- usage of $varname in many daemons
All daemons are updated to use single varname "$vrf_name".
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
Problem reported that when a "neighbor x.x.x.x route-map FOO in"
set a next-hop value, that modified next-hop value was also sent
to eBGP peers. This is incorrect since bgp is expected to set
next-hop to self when sending to eBGP peers unless third party
next-hop on a shared segment is true. This fix modifies the
behavior to stop sending the modified next-hop to eBGP peers
if the route-map was applied inbound on another peer.
Ticket: CM-26025
Signed-off-by: Don Slice <dslice@cumulusnetworks.com>
The newly added PEER_RMAP_TYPE_AGGREGATE flag is setup to
be the 9th bit:
But the flag we are putting it into:
uint8_t rmap_type;
is 8 bits. Adjust the size.
Found by Coverity SA Scan
Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>
The "abort_if_not_found" parameter of nb_running_get_entry()
should be set to true only when this function is called during the
NB_EV_APPLY phase of a northbound callback. Failure to respect this
can lead to crashes when multiple configuration changes are being
committed at the same time.
Signed-off-by: Renato Westphal <renato@opensourcerouting.org>
RFC 4271 sec 6.3 p33, In the case of a BGP_NEXTHOP attribute with an
incorrect value, FRR is supposed to send a notification
and include 'Corresponding type, length and value of the NEXT_HOP
attribute in the notification data.
Fixes: #4997
Signed-off-by: Nikos <ntriantafillis@gmail.com>
Signed-off-by: Donald Sharp <sharpd@cumulusnetworks.com>
[7.2 version]
When processing route updates from the dataplane, we were
terminating the checking of nexthops prematurely, and we could
miss meaningful changes.
Signed-off-by: Mark Stapp <mjs@voltanet.io>
