bgpd: Intern default-originate attributes to avoid use-after-free
When we receive a default route from a peer and we originate default route
using `neighbor default-originate`, we do not track of struct attr we use,
and when we do `no neighbor default-originate` we withdraw our generated
default route, but we announce default-route from the peer.
After we do this, we unintern aspath (which was used for default-originate),
BUT it was used also for peer's default route we received.
And here we have a use-after-free crash, because bgp_process_main_one()
reaps old paths that are marked as BGP_PATH_REMOVED with aspath->refcnt > 0,
but here it's 0.
```
0 0x55c24bbcd022 in aspath_key_make bgpd/bgp_aspath.c:2070
1 0x55c24b8f1140 in attrhash_key_make bgpd/bgp_attr.c:777
2 0x7f52322e66c9 in hash_release lib/hash.c:220
3 0x55c24b8f6017 in bgp_attr_unintern bgpd/bgp_attr.c:1271
4 0x55c24ba0acaa in bgp_path_info_free_with_caller bgpd/bgp_route.c:283
5 0x55c24ba0a7de in bgp_path_info_unlock bgpd/bgp_route.c:309
6 0x55c24ba0af6d in bgp_path_info_reap bgpd/bgp_route.c:426
7 0x55c24ba17b9a in bgp_process_main_one bgpd/bgp_route.c:3333
8 0x55c24ba18a1d in bgp_process_wq bgpd/bgp_route.c:3425
9 0x7f52323c2cd5 in work_queue_run lib/workqueue.c:282
10 0x7f52323aab92 in thread_call lib/thread.c:2006
11 0x7f5232300dc7 in frr_run lib/libfrr.c:1198
12 0x55c24b8ea792 in main bgpd/bgp_main.c:520
13 0x7f5231c3a082 in __libc_start_main ../csu/libc-start.c:308
14 0x55c24b8ef0bd in _start (/usr/lib/frr/bgpd+0x2c90bd)
```
tools: Fetch more commits for commitlint to validate the commit
Note: It's necessary that you specify the fetch-depth argument to
actions/checkout@v2 step. By default they fetch only latest commit of the
branch, but we need more commits since we validate a range of commit messages.
anlan_cs [Mon, 6 Feb 2023 01:27:05 +0000 (09:27 +0800)]
bgpd: fix use-after-free crash for evpn
```
anlan(config-router-af)# vni 33
anlan(config-router-af-vni)# route-target both 44:55
anlan(config-router-af-vni)# no route-target both 44:55
vtysh: error reading from bgpd: Resource temporarily unavailable (11)Warning: closing connection to bgpd because of an I/O error!
```
When `bgp_evpn_vni_rt_cmd` deals with "both" type, it wrongly created
only one node ( should be two nodes ) for lists of both `vpn->import_rtl` and
`vpn->export_rtl`. At this time, the two lists are already wrong.
In `no route-target both RT`, it will free the single node from lists of both
`vpn->import_rtl` and `vpn->export_rtl`. After freed from `vpn->import_rtl`,
it is "use-after-free" at the time of freeing it from `vpn->export_rtl`.
It causes crash sometimes, or other unexpected behaviours.
This issue is introduced by commit `3b7e8d`, which have adjusted both
`bgp_evpn_vni_rt_cmd` and `bgp_evpn_vrf_rt_cmd`.
Since `bgp_evpn_vrf_rt_cmd/no_bgp_evpn_vrf_rt_cmd` works well again
unintentionally with commit `7022da`, only `bgp_evpn_vni_rt_cmd` needs to
modify - add two nodes for "both" type and some explicit comments for this
special case of "both" type.
Donald Sharp [Thu, 2 Feb 2023 19:13:12 +0000 (14:13 -0500)]
bgpd: Don't try to recursively hold peer io mutex
BGP was modified in a0b937de428e14e869b8541f0b7810113d619c2e
to grab the peer->io_mtx before validating the header to ensure
that the input Queue was not being modified by anyone else at that
moment in time. Unfortunately validate_header can detect a problem
and attempt to relock the mutex, which deadlocks. This deadlock in
the bgp_io pthread is the lone deadlock at first, eventually though
bgp attempts to write another packet to the peer( say when the
it's time to send the next packet ) and the main pthread of bgpd
becomes deadlocked and then the whole bgpd process is stuck at that
point in time leaving us dead in the water.
The point of locking the mutex earlier was to ensure that the input
Queue wasn't being modified by anyone else, (Say reading off it )
as that we wanted to ensure that we don't hold more packets then necessary.
Let's grab the mutex long enough to look at the input Q size, this
ensure that we have room and then we can validate_header and do the right
thing from there. We'll need to lock the mutex when we actually move it
into the input Q as well.
Fixes: #12725 Signed-off-by: Donald Sharp <sharpd@nvidia.com>
Donald Sharp [Thu, 2 Feb 2023 15:40:07 +0000 (10:40 -0500)]
bgpd: Convert evpn output to not pretty print json
Commit: 3cdb03fba7b40240fb38469a12b7b05a11043e09
changed the vty_json output to not be pretty printing.
The previous commit in the tree added vty_json_no_pretty
let's use that instead
Donald Sharp [Thu, 2 Feb 2023 15:28:19 +0000 (10:28 -0500)]
lib, bgpd: Add ability to specify that some json output should not be pretty
Initial commit: 23b2a7ef524c9fe083b217c7f6ebaec0effc8f52
changed the json output of `show bgp <afi> <safi> json` to
not have pretty print because when under a situation where
there are a bunch of routes with a large scale ecmp show
output was taking forever and this commit cut 2 minutes out
of vtysh run time.
When upgrading to latest version the long run time was noticed
due to testing. Let's add back this functionality such that
FRR can have reduced run times with vtysh when it's really
needed.
Donatas Abraitis [Tue, 24 Jan 2023 08:32:13 +0000 (10:32 +0200)]
bgpd: Set attr to NULL when passing NLRI_UPDATE with treat-as-withdraw
Before this patch, we always passed `struct attr` for NLRI_UPDATE, but if we
have a situation with treat-as-withdraw (for example: malformed attribute, or
using a command like `neighbor path-attribute treat-as-withdraw`) the route
MUST be withdrawn form the BGP table.
Hence, we MUST pass attr as NULL, in this case we already have this check
under NLRI_ATTR_ARG() macro, just reuse it properly.
Donald Sharp [Mon, 30 Jan 2023 15:12:06 +0000 (10:12 -0500)]
*: Add -Wswitch-enum to build
The -Wswitch-enum will allow the compiler to warn us
when a developer creates a switch over a enum and is
using `default:` when they should be iterating over
every enum
Donald Sharp [Fri, 27 Jan 2023 13:10:08 +0000 (08:10 -0500)]
bgpd: ecommunity_token_rt6 is not handled
The function ecommunity_str2com_internal appears to want to handle
the ecommunity_token_rt6 enum but skips over it. Commit 9a659715dfcb6c0b1e3ef8004b6c9d14c55f2081 tried to add this but I really
don't see how this is going to behave correctly. Add the
ecommunity_token_rt6 case to the switch statement so it is handled
appropriately?
Donald Sharp [Fri, 27 Jan 2023 13:10:08 +0000 (08:10 -0500)]
bgpd: ecommunity_token_rt6 is not handled
The function ecommunity_str2com_internal appears to want to handle
the ecommunity_token_rt6 enum but skips over it. Commit 9a659715dfcb6c0b1e3ef8004b6c9d14c55f2081 tried to add this but I really
don't see how this is going to behave correctly. Add the
ecommunity_token_rt6 case to the switch statement so it is handled
appropriately?
Donald Sharp [Thu, 19 Jan 2023 20:46:31 +0000 (15:46 -0500)]
tests: zebra_rib remove a sleep
The test was sometimes failing around the sleep(4) for
waiting for the routes to be installed. Instead of blindly
sleeping let's check to see that the routes are actually
there in zebra and then continue on.
Donald Sharp [Mon, 30 Jan 2023 18:53:44 +0000 (13:53 -0500)]
zebra: Send nht resolved entry up to concerned protocols in all cases
There existed the idea, from Volta, that a nexthop group would not have
the same nexthops installed -vs- what FRR actually sent down. The
dplane would notify you.
The flag ROUTE_ENTRY_USE_FIB_NHG flag was being used
to control which set was being sent up to concerned parties
in nexthop tracking. Put this flag behind the wall and
do not necessarily set it when we receive a data plane
notification about a route being installed or not.
Fixes: #12706 Signed-off-by: Donald Sharp <sharpd@nvidia.com>
Donald Sharp [Mon, 30 Jan 2023 21:02:23 +0000 (16:02 -0500)]
bgpd: bgp_update and bgp_withdraw never return failures
These two functions always return 0. As such any and all
tests against this make no sense. Remove the return 0
to a void and follow the chain, logically, to remove all
the dead code.
Donald Sharp [Sat, 28 Jan 2023 14:32:34 +0000 (09:32 -0500)]
babeld: During intf startup, ignore address already in use
When listening on a multicast group. No need to actually
fail the operation when it's already being used.
Let's not treat the Address already in use error message
as one that is stopping everything from working. Especially
since multiple interface events cause this to happen.
Without this, if config is read in before full connection
to zebra, babel will never establish neighbors.
Trey Aspelund [Wed, 25 Jan 2023 18:25:20 +0000 (13:25 -0500)]
bgpd: move tunnel-ip comparison into handler
Moves the old/new IP comparison into handle_tunnel_ip_change instead of
expecting the caller to do the check on their own.
Also changes handle_tunnel_ip_change to return void since it only ever
returned 0 in all cases.
Trey Aspelund [Wed, 25 Jan 2023 18:07:43 +0000 (13:07 -0500)]
bgpd: only unimport routes if tunnel-ip changes
When processing a new local VNI, we were always walking the global EVPN
table to look for routes that needed to be removed due to a martian
nexthop change (specifically a tunnel-ip change).
Since the martian TIP table is global (all VNIs) + the walk is also in
the global table (all VNIs), we can trust that any new TIP from any VNI
would result in routes getting removed from the global table and
unimported from all live (L2)VNIs.
i.e.
The only time this update is actionable is if we are adding/removing an
IP from the martian TIP table, and we do not need to walk the table for
normal refcount adjustments.
projects / mirror / frr.git / log