Add a hash_clean_and_free() function as well as convert
the code to use it. This function also takes a double
pointer to the hash to set it NULL. Also it cleanly
does nothing if the pointer is NULL( as a bunch of
code tested for ).
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
Passing a pre-formatted buffer in these places needs a `"%s"` in front
so it doesn't get formatted twice.
Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
If we got inside the condition of `vrfp->status == VRF_ACTIVE` then
don't make the same check again.
Found by Coverity Scan (CID 1519760)
Signed-off-by: Rafael Zalamena <rzalamena@opensourcerouting.org>
We must disable the vrf before we start terminating interfaces.
On termination, we free the 'zebra_if' struct from the interface ->info
pointer. We rely on that for subsystems like vxlan for cleanup when
shutting down.
'''
==497406== Invalid read of size 8
==497406== at 0x47E70A: zebra_evpn_del (zebra_evpn.c:1103)
==497406== by 0x47F004: zebra_evpn_cleanup_all (zebra_evpn.c:1363)
==497406== by 0x4F2404: zebra_evpn_vxlan_cleanup_all (zebra_vxlan.c:1158)
==497406== by 0x4917041: hash_iterate (hash.c:267)
==497406== by 0x4F25E2: zebra_vxlan_cleanup_tables (zebra_vxlan.c:5676)
==497406== by 0x4D52EC: zebra_vrf_disable (zebra_vrf.c:209)
==497406== by 0x49A247F: vrf_disable (vrf.c:340)
==497406== by 0x49A2521: vrf_delete (vrf.c:245)
==497406== by 0x49A2E2B: vrf_terminate_single (vrf.c:533)
==497406== by 0x49A2D8F: vrf_terminate (vrf.c:561)
==497406== by 0x441240: sigint (main.c:192)
==497406== by 0x4981F6D: frr_sigevent_process (sigevent.c:130)
==497406== Address 0x6d68c68 is 200 bytes inside a block of size 272 free'd
==497406== at 0x48470E4: free (vg_replace_malloc.c:872)
==497406== by 0x4942CF0: qfree (memory.c:141)
==497406== by 0x49196A9: if_delete (if.c:293)
==497406== by 0x491C54C: if_terminate (if.c:1031)
==497406== by 0x49A2E22: vrf_terminate_single (vrf.c:532)
==497406== by 0x49A2D8F: vrf_terminate (vrf.c:561)
==497406== by 0x441240: sigint (main.c:192)
==497406== by 0x4981F6D: frr_sigevent_process (sigevent.c:130)
==497406== by 0x499A5F0: thread_fetch (thread.c:1775)
==497406== by 0x492850E: frr_run (libfrr.c:1197)
==497406== by 0x441746: main (main.c:476)
==497406== Block was alloc'd at
==497406== at 0x4849464: calloc (vg_replace_malloc.c:1328)
==497406== by 0x49429A5: qcalloc (memory.c:116)
==497406== by 0x491D971: if_new (if.c:174)
==497406== by 0x491ACC8: if_create_name (if.c:228)
==497406== by 0x491ABEB: if_get_by_name (if.c:613)
==497406== by 0x427052: netlink_interface (if_netlink.c:1178)
==497406== by 0x43BC18: netlink_parse_info (kernel_netlink.c:1188)
==497406== by 0x4266D7: interface_lookup_netlink (if_netlink.c:1288)
==497406== by 0x42B634: interface_list (if_netlink.c:2368)
==497406== by 0x4ABF83: zebra_ns_enable (zebra_ns.c:127)
==497406== by 0x4AC17E: zebra_ns_init (zebra_ns.c:216)
==497406== by 0x44166C: main (main.c:408)
'''
Signed-off-by: Stephen Worley <sworley@nvidia.com>
Currently, it is possible to rename the default VRF either by passing
`-o` option to zebra or by creating a file in `/var/run/netns` and
binding it to `/proc/self/ns/net`.
In both cases, only zebra knows about the rename and other daemons learn
about it only after they connect to zebra. This is a problem, because
daemons may read their config before they connect to zebra. To handle
this rename after the config is read, we have some special code in every
single daemon, which is not very bad but not desirable in my opinion.
But things are getting worse when we need to handle this in northbound
layer as we have to manually rewrite the config nodes. This approach is
already hacky, but still works as every daemon handles its own NB
structures. But it is completely incompatible with the central
management daemon architecture we are aiming for, as mgmtd doesn't even
have a connection with zebra to learn from it. And it shouldn't have it,
because operational state changes should never affect configuration.
To solve the problem and simplify the code, I propose to expand the `-o`
option to all daemons. By using the startup option, we let daemons know
about the rename before they read their configs so we don't need any
special code to deal with it. There's an easy way to pass the option to
all daemons by using `frr_global_options` variable.
Unfortunately, the second way of renaming by creating a file in
`/var/run/netns` is incompatible with the new mgmtd architecture.
Theoretically, we could force daemons to read their configs only after
they connect to zebra, but it means adding even more code to handle a
very specific use-case. And anyway this won't work for mgmtd as it
doesn't have a connection with zebra. So I had to remove this option.
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
If we're exiting before we finished initializing, we can end up trying
to shut down a NULL vrf here.
Signed-off-by: David Lamparter <equinox@opensourcerouting.org>
This is needed for the following two reasons:
1. To be able to remove the northbound HACK in if_update_to_new_vrf. It
is totally wrong to rewrite the configuration datastore when some
operational state changes. It is a hard blocker for storing a
configuration data in a management daemon which knows nothing about
the operational state.
2. To allow changing the VRF of the interface using FRR CLI or any other
frontend in the future. If the VRF is a part of the key, it can't be
changed. If the VRF is a simple leaf, it becomes possible to change
it and thus move the interface between VRFs. For now I mark the leaf
as a "config false" as it's not yet possible to control it from FRR.
But we can't simply remove the VRF from the key, because it is needed to
distinguish interfaces when using netns based VRFs, as it is possible to
have multiple interfaces with the same name in different namespaces. To
handle this, I came up with an idea to store both VRF and an interface
name in the "name" leaf using the pattern "vrfname:ifname". For example,
if there's an interface "eth0" in VRF "red" then its "name" leaf will be
"red:eth0".
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
Since f60a1188 we store a pointer to the VRF in the interface structure.
There's no need anymore to store a separate vrf_id field.
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
Currently, we automatically delete an inactive VRF when its last
interface is deleted. This code introduces a couple of crashes because
of the following problems:
- vrf_delete is called before calling if_del hook, so daemons may try to
dereference an ifp->vrf pointer which is freed
- in if_terminate, we continue to use the VRF in the loop condition
after the last interface is deleted
This check is needed only when the interface is deleted by the user,
because if the interface is deleted by the system, VRF must still exist
in the system. Move the check to appropriate places to fix crashes.
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
If the VRF is not enabled, if_terminate deletes the VRF after the last
interface is removed from it. Therefore daemons crash on the subsequent
call to vrf_delete. We should call vrf_delete only for enabled VRFs.
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
When the netns is deleted, we should always clear the vrf->ns_ctxt
pointer. Currently, it is not cleared when there are interfaces in the
netns at the time of deletion.
If the netns is re-created, zebra crashes because it tries to use the
stale pointer.
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
It allows FRR to read the interface config even when the necessary VRFs
are not yet created and interfaces are in "wrong" VRFs. Currently, such
config is rejected.
For VRF-lite backend, we don't care at all about the VRF of the inactive
interface. When the interface is created in the OS and becomes active,
we always use its actual VRF instead of the configured one. So there's
no need to reject the config.
For netns backend, we may have multiple interfaces with the same name in
different VRFs. So we care about the VRF of inactive interfaces. And we
must allow to preconfigure the interface in a VRF even before it is
moved to the corresponding netns. From now on, we allow to create
multiple configs for the same interface name in different VRFs and
the necessary config is applied once the OS interface is moved to the
corresponding netns.
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
When something is used only from zebra and part of its description is
"should be called from zebra only" then it belongs to zebra, not lib.
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
vrf_name_to_id() returned VRF_DEFAULT when the vrf name was
unknown, hiding errors. Per community recommendation, vrf_name_to_id()
is now removed and the few callers now use vrf_lookup_by_name()
directly.
Signed-off-by: G. Paul Ziemba <paulz@labn.net>
"[no] netns NAME" commands are part of the lib, but they are actually
zebra-only:
- they are using vrf_netns_handler_create and its description clearly
says that it "should be called from zebra only"
- vtysh sends these commands only to zebra
- only zebra outputs the netns related config
- zebra notifies other daemons about netns attachment
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
If we have the following configuration:
```
vrf red
smth
exit-vrf
!
interface red vrf red
smth
```
And we delete the VRF using "no vrf red" command, we end up with:
```
interface red
smth
```
Interface config is preserved but moved to the default VRF.
This is not an expected behavior. We should remove the interface config
when the VRF is deleted.
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
Always terminate default VRF last during FRR shutdown.
On shutdown we were simply looping over the RB tree and terminating
VRFs from the ROOT. This is not guaranteed to be the default last ever.
Instead switch to RB_SAFE and skip the default VRF till the very end.
Signed-off-by: Stephen Worley <sworley@nvidia.com>
When the VRF node is exited using "exit" or "quit", there's still a VRF
pointer stored in the vty context. If you try to configure some router
related command, it will be applied to the previous VRF instead of the
default VRF. For example:
```
(config)# vrf test
(config-vrf)# ip router-id 1.1.1.1
(config-vrf)# do show run
...
!
vrf test
ip router-id 1.1.1.1
exit-vrf
!
...
(config-vrf)# exit
(config)# ip router-id 2.2.2.2
(config)# do show run
...
!
vrf test
ip router-id 2.2.2.2
exit-vrf
!
...
```
`vrf-exit` works correctly, because it stores a pointer to the default
VRF into the vty context (but weirdly keeping the VRF_NODE instead of
changing it to CONFIG_NODE).
Instead of relying on the behavior of exit function, always use the
default VRF when in CONFIG_NODE.
Another problem is missing `VTY_CHECK_CONTEXT`. If someone deletes the
VRF in which node the user enters the command, then zebra applies the
command to the default VRF instead of throwing an error.
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
The backoff code assumed that yang operations always completed quickly.
It checked for > 100 YANG modeled commands happening in under 1 second
to enable batching. If 100 yang modeled commands always take longer than
1 second batching is never enabled. This is the exact opposite of what
we want to happen since batching speeds the operations up.
Here are the results for libyang2 code without and with batching.
| action | 1K rts | 2K rts | 1K rts | 2K rts | 20k rts |
| | nobatch | nobatch | batch | batch | batch |
| Add IPv4 | .881 | 1.28 | .703 | 1.04 | 8.16 |
| Add Same IPv4 | 28.7 | 113 | .590 | .860 | 6.09 |
| Rem 1/2 IPv4 | .376 | .442 | .379 | .435 | 1.44 |
| Add Same IPv4 | 28.7 | 113 | .576 | .841 | 6.02 |
| Rem All IPv4 | 17.4 | 71.8 | .559 | .813 | 5.57 |
(IPv6 numbers are basically the same as iPv4, a couple percent slower)
Clearly we need this. Please note the growth (1K to 2K) w/o batching is
non-linear and 100 times slower than batched.
Notes on code: The use of the new `nb_cli_apply_changes_clear_pending`
is to commit any pending changes (including the current one). This is
done when the code would not correctly handle a single diff that
included the current changes with possible following changes. For
example, a "no" command followed by a new value to replace it would be
merged into a change, and the code would not deal well with that. A good
example of this is BGP neighbor peer-group changing. The other use is
after entering a router level (e.g., "router bgp") where the follow-on
command handlers expect that router object to now exists. The code
eventually needs to be cleaned up to not fail in these cases, but that
is for future NB cleanup.
Signed-off-by: Christian Hopps <chopps@labn.net>
There are two possible use-cases for the `vrf_bind` function:
- bind socket to an interface in a vrf
- bind socket to a vrf device
For the former case, there's one problem - success is returned when the
interface is not found. In that case, the socket is left unbound without
throwing an error.
For the latter case, there are multiple possible problems:
- If the name is not set, then the socket is left unbound (zebra, vrrp).
- If the name is "default" and there's an interface with that name in the
default VRF, then the socket is bound to that interface.
- In most daemons, if the router is configured before the VRF is actually
created, we're trying to open and bind the socket right after the
daemon receives a VRF registration from zebra. We may not receive the
VRF-interface registration from zebra yet at that point. Therefore,
`if_lookup_by_name` fails, and the socket is left unbound.
This commit fixes all the issues and updates the function description.
Suggested-by: Pat Ruddy <pat@voltanet.io>
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
Compile with v2.0.0 tag of `libyang2` branch of:
https://github.com/CESNET/libyang
staticd init load time of 10k routes now 6s vs ly1 time of 150s
Signed-off-by: Christian Hopps <chopps@labn.net>
This is to fix the crash reproduced by the following steps:
* ip link add red type vrf table 1
Creates VRF.
* vtysh -c "conf" -c "vrf red"
Creates VRF NB node and marks VRF as configured.
* ip route 1.1.1.0/24 2.2.2.2 vrf red
* no ip route 1.1.1.0/24 2.2.2.2 vrf red
(or similar l3vni set/unset in zebra)
Marks VRF as NOT configured.
* ip link del red
VRF is deleted, because it is marked as not configured, but NB node
stays.
Subsequent attempt to configure something in the VRF leads to a crash
because of the stale pointer in NB layer.
Fixes#8357.
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
Back when I put this together in 2015, ISO C11 was still reasonably new
and we couldn't require it just yet. Without ISO C11, there is no
"good" way (only bad hacks) to require a semicolon after a macro that
ends with a function definition. And if you added one anyway, you'd get
"spurious semicolon" warnings on some compilers...
With C11, `_Static_assert()` at the end of a macro will make it so that
the semicolon is properly required, consumed, and not warned about.
Consistently requiring semicolons after "file-level" macros matches
Linux kernel coding style and helps some editors against mis-syntax'ing
these macros.
Signed-off-by: David Lamparter <equinox@diac24.net>
Neither tabs nor newlines are acceptable in syslog messages. They also
break line-based parsing of file logs.
Signed-off-by: David Lamparter <equinox@diac24.net>
The VRF must be marked as configured when user enters "vrf NAME" command.
Otherwise, the following problem occurs:
`ip link add red type vrf table 1`
VRF structure is allocated.
`vtysh -c "conf t" -c "vrf red"`
`lib_vrf_create` is called, and pointer to the VRF structure is stored
to the nb_config_entry.
`ip link del red`
VRF structure is freed (because it is not marked as configured), but
the pointer is still stored in the nb_config_entry.
`vtysh -c "conf t" -c "no vrf red"`
Nothing happens, because VRF structure doesn't exist. It means that
`lib_vrf_destroy` is not called, and nb_config_entry still exists in
the running config with incorrect pointer.
`ip link add red type vrf table 1`
New VRF structure is allocated.
`vtysh -c "conf t" -c "vrf red"`
`lib_vrf_create` is NOT called, because the nb_config_entry for that
VRF name still exists in the running config.
After that all NB commands for this VRF will use incorrect pointer to
the freed VRF structure.
Signed-off-by: Igor Ryzhov <iryzhov@nfware.com>
Description: When we get a new vrf add and vrf with same name, but different vrf-id already
exists in the database, we should treat vrf add as update.
This happens mostly when there are lots of vrf and other configuration being replayed.
There may be a stale vrf delete followed by new vrf add. This
can cause timing race condition where vrf delete could be missed and
further same vrf add would get rejected instead of treating last arrived
vrf add as update.
Treat vrf add for existing vrf as update.
Implicitly disable this VRF to cleanup routes and other functions as part of vrf disable.
Update vrf_id for the vrf and update vrf_id tree.
Re-enable VRF so that all routes are freshly installed.
Above 3 steps are mandatory since it can happen that with config reload
stale routes which are installed in vrf-1 table might contain routes from
older vrf-0 table which might have got deleted due to missing vrf-0 in new configuration.
Signed-off-by: sudhanshukumar22 <sudhanshu.kumar@broadcom.com>
Code was added in the past to support a value of VRF_DEFAULT different
from 0. This option was abandoned, the default vrf id is always 0.
Remove this code, this will simplify the code and improve performance
(use a constant value instead of a function that performs tests).
Signed-off-by: Christophe Gouault <christophe.gouault@6wind.com>
vrf_id_to_name() looks up in a RB_TREE to find the VRF entry, then
reads the name.
Avoid it for VRF_DEFAULT, which always exists and for which the
translation is straightforward.
Signed-off-by: Christophe Gouault <christophe.gouault@6wind.com>
The vrf_get function is called throughout the code base
so much so that when you turn on vrf debugging it eclipses
everything else to a degree that is completely unreasonable.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
The vrf name was not being printed out in some vrf debugs. Add
this data in so people don't have to remember the vrf id.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
When the nexthop-groups were added to FRR for some
reason the call to nexthop_group_disable_vrf was
not added although it was written.
Add it in.
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
to be able to retrieve the network namespace identifier for each
namespace, the ns id is stored in each ns context. For default
namespace, the netns id is the same as that value.
Signed-off-by: Philippe Guibert <philippe.guibert@6wind.com>
Revert "zebra: support for macvlan interfaces"
This reverts commit bf69e212fd.
Revert "doc: add some documentation about bgp evpn netns support"
This reverts commit 89b97c33d7.
Revert "zebra: dynamically detect vxlan link interfaces in other netns"
This reverts commit de0ebb2540.
Revert "bgpd: sanity check when updating nexthop from bgp to zebra"
This reverts commit ee9633ed87.
Revert "lib, zebra: reuse and adapt ns_list walk functionality"
This reverts commit c4d466c830.
Revert "zebra: local mac entries populated in correct netnamespace"
This reverts commit 4042454891.
Revert "zebra: when parsing local entry against dad, retrieve config"
This reverts commit 3acc394bc5.
Revert "bgpd: evpn nexthop can be changed by default"
This reverts commit a2342a2412.
Revert "zebra: zvni_map_to_vlan() adaptation for all namespaces"
This reverts commit db81d18647.
Revert "zebra: add ns_id attribute to mac structure"
This reverts commit 388d5b438e.
Revert "zebra: bridge layer2 information records ns_id where bridge is"
This reverts commit b5b453a2d6.
Revert "zebra, lib: new API to get absolute netns val from relative netns val"
This reverts commit b6ebab34f6.
Revert "zebra, lib: store relative default ns id in each namespace"
This reverts commit 9d3555e06c.
Revert "zebra, lib: add an internal API to get relative default nsid in other ns"
This reverts commit 97c9e7533b.
Revert "zebra: map vxlan interface to bridge interface with correct ns id"
This reverts commit 7c990878f2.
Revert "zebra: fdb and neighbor table are read for all zns"
This reverts commit f8ed2c5420.
Revert "zebra: zvni_map_to_svi() adaptation for other network namespaces"
This reverts commit 2a9dccb647.
Revert "zebra: display interface slave type"
This reverts commit fc3141393a.
Revert "zebra: zvni_from_svi() adaptation for other network namespaces"
This reverts commit 6fe516bd4b.
Revert "zebra: importation of bgp evpn rt5 from vni with other netns"
This reverts commit 28254125d0.
Revert "lib, zebra: update interface name at netlink creation"
This reverts commit 1f7a68a2ff.
Signed-off-by: Pat Ruddy <pat@voltanet.io>
When using the default CLI mode, the northbound layer needs to create
a separate transaction to process each YANG-modeled command since
they are supposed to be applied immediately (there's no candidate
configuration nor the "commit" command like in the transactional
CLI). The problem is that configuration transactions have an overhead
associated to them, in big part because of the use of some heavy
libyang functions like `lyd_validate()` and `lyd_diff()`. As of
now this overhead is substantial and doesn't scale well when large
numbers of transactions need to be performed in sequence.
As an example, loading 50k prefix-lists using a single transaction
takes about 2 seconds on a modern CPU. Loading the same 50k
prefix-lists using 50k transactions can take more than an hour
to complete (which is unacceptable by any standard). To fix this
problem, some heavy optimization work needs to be done on libyang and
on the FRR northbound itself too (e.g. perform partial configuration
diffs whenever possible). This, however, should be a long term
effort since these optimizations shouldn't be trivial to implement
and we're far from having the performance numbers we need.
In the meanwhile, this commit introduces a simple but efficient
workaround to alleviate the issue. In short, a new back-off timer
was introduced in the CLI to monitor and detect when too many
YANG-modeled commands are being received at the same time. When
a certain threshold is reached (100 YANG-modeled commands within
one second), the northbound starts to group all subsequent commands
into a single large transaction, which allows them to be processed
much faster (e.g. seconds and not hours). It's essentially a
protection mechanism that creates dynamically-sized transactions
when necessary to prevent performance issues from happening. This
mechanism is enabled both when parsing configuration files and when
reading commands from a terminal.
The downside of this optimization is that, if several YANG-modeled
commands are grouped into the same transaction and at least one of
them fails, the whole transaction is rejected. This is undesirable
since users don't expect transactional behavior when that's not
enabled explicitly. To minimize this issue, the CLI will log all
commands that were rejected whenever that happens, to make the
user aware of what happened and have enough information to fix
the problem. Commands that fail due to parsing errors or CLI-level
validations in general are rejected separately.
Again, this proposed workaround is intended to be temporary. The
goal is to provided a quick fix to issues like #6658 while we work
on better long-term solutions.
Signed-off-by: Renato Westphal <renato@opensourcerouting.org>
DEFPY_YANG will allow the CLI to identify which commands are
YANG-modeled or not before executing them. This is going to be
useful for the upcoming configuration back-off timer work that
needs to commit pending configuration changes before executing a
command that isn't YANG-modeled.
Signed-off-by: Renato Westphal <renato@opensourcerouting.org>
