| Age | Commit message (Collapse) | Author |
|---|
|
Fix the capability numbers for the next few releases already. While not
yet in use, the number is reserved and stable.
|
|
We use clang-format, but the result depends on the version of the tool.
The version to be used, is the one from the github action. You can use
"tools/clang-format-container.sh" to run that version in a container.
Now that Fedora 40 is released, bump the Fedora version (and the used
clang-format version) from Fedora 39 (17.0.6-2.fc39) to Fedora 40
(18.1.1-1.fc40). Luckily, our source code formats the same for both.
|
|
When the kernel sends a ECMP route update with just the deleted nexthop,
the nexthop will have no associated weight, and its flags may indicate
that it is dead:
route_update: RTM_DELROUTE
new route:
inet6 default table main type unicast <DEAD,>
scope global priority 0x400 protocol 0x9
nexthop via fe80::b226:28ff:fe62:8841 dev port4 <dead,>
old route:
inet6 default table main type unicast
scope global priority 0x400 protocol 0x9
nexthop via fe80::b226:28ff:fe62:8841 dev port4 weight 0 <>
nexthop via fe80::fa8e:a1ff:fee0:8344 dev port49 weight 0 <>
nexthop via fe80::b226:28ff:fe62:d400 dev port3 weight 0 <>
nexthop via fe80::fa8e:a1ff:fee0:8349 dev port54 weight 0 <>
Since we are comparing the nexthops strictly with all attributes, we can
never match the deleted nexthop. This causes libnl to fail to remove the
deleted nexthop from the route, and consequently send out a nop-update
and a desync of the route in the cache and in the kernel.
Fix this by ignoring NH_ATTR_FLAGS (0x1) and NH_ATTR_WEIGHT (0x2) when
comparing nexthops to properly match the deleted one.
Fixes: 29b71371e764 ("route cache: Fix handling of ipv6 multipath routes")
Signed-off-by: Jonas Gorski <jonas.gorski@bisdn.de>
https://github.com/thom311/libnl/pull/382
|
|
https://github.com/thom311/libnl/pull/378
|
|
No "goto errout".
|
|
- mngr->cm_flags must be set together when assigning the socket.
Otherwise, a `goto errout` in the middle will cause a leak.
- normalize the flags variable to not contain unexpected values.
- NL_ALLOCATED_SYNC_SOCK is private API. No need to expose that to
public headers.
Fixes: 1dbdc30a6b9f ('cache: allow to allocate cache manager with custom refill socket')
|
|
Cache managers use two sockets: one for cache refill operation,
and another one for notifications.
In order to simulate NETLINK events by reading data from files,
we need to be able to overwrite callbacks for both sockets.
This new function allows us to set up refill socket any way we want.
It does have requirement that the refill socket be blocking.
|
|
While libnl3 is a C library (and itself can only be built using a C
compiler), the public headers are supposed to also work with C++.
Add a test for that.
Unfortunately, this test does not cover whether all symbols are
correctly marked as "extern "C"" for linkage.
|
|
|
|
deprecated headers
Header files are part of a stable API. Warning about using deprecated
API is cumbersome, albeit often useful to force the user to migrate
away from the API.
But it also requires that the unit test disables "-Wcpp". That is
problematic on its own, because we don't want to disable any warnings.
As we only have 4 such header files, instead guard the warnings with
a
#ifndef _NL_NO_WARN_DEPRECATED_HEADER
|
|
printf "#include <$(echo "include/netlink/addr.h" | sed 's|.*\<include/netlink/|netlink/|')>\nint main(int argc, char **argv) { return 0; }" > include/netlink/addr.h.build-headers-test.c
gcc -DHAVE_CONFIG_H -I. -I./include -O2 -Wall -Wextra -Werror -Wall -Werror -Wno-error=cpp -I./include -I./include -c -o include/netlink/addr.h.build-headers-test.o include/netlink/addr.h.build-headers-test.c
include/netlink/addr.h.build-headers-test.c: In function ‘main’:
include/netlink/addr.h.build-headers-test.c:2:14: error: unused parameter ‘argc’ [-Werror=unused-parameter]
2 | int main(int argc, char **argv) { return 0; }
| ~~~~^~~~
include/netlink/addr.h.build-headers-test.c:2:27: error: unused parameter ‘argv’ [-Werror=unused-parameter]
2 | int main(int argc, char **argv) { return 0; }
| ~~~~~~~^~~~
cc1: all warnings being treated as errors
Fixes: d9a1e0ce9c95 ('build: add "check-local-build-headers" test target to build public headers')
|
|
rtnl_netem_set_delay_distribution()
CC lib/route/qdisc/libnl_route_3_la-netem.lo
lib/route/qdisc/netem.c: In function 'rtnl_netem_set_delay_distribution':
lib/route/qdisc/netem.c:975:39: error: 'calloc' sizes specified with 'sizeof' in the earlier argument and not in the later argument [-Werror=calloc-transposed-args]
975 | line = (char *) calloc(sizeof(char), len + 1);
| ^~~~
lib/route/qdisc/netem.c:975:39: note: earlier argument should specify number of elements, later size of each element
|
|
When calling the callback v2 for objects that were updated, we pass the
update ("obj") instead of the updated object ("old") as new.
Presumably this wasn't intended, so pass the updated object as new.
This avoids weird updates where the new object is significantly smaller
than the old one. E.g. for IPv6 multipath route updates, old would be
the full route with all nexthops, while new would be a partial route
with only the added/removed nexthop.
Fixes: 66d032ad443a ("cache_mngr: add include callback v2")
Signed-off-by: Jonas Gorski <jonas.gorski@bisdn.de>
https://github.com/thom311/libnl/pull/381
|
|
n needs to be uint32_t to fit left shift by 22 bits
https://github.com/thom311/libnl/pull/379
|
|
|
|
In this case, only test it with clang. It seems not worth building
everything twice toggling only this option.
|
|
Otherwise, whether libnl-3.so exports nl_debug_dp depends on NL_DEBUG.
That is ugly. It also breaks the linker checking the symbol versioning
file with the "--no-undefined-version" flag.
Instead, always define it. It's small anyway.
Reported-by: lch361 <lch361@skiff.com>
See-also: https://github.com/thom311/libnl/pull/375
|
|
Conditionally defining to nothing, means that the compiler doesn't see
the print statement without NL_DEBUG. In turn, we lack checking of the
statement by the compiler.
Instead, add an "if (NL_DEBUG)" around it. Since NL_DEBUG is a constant, the
compiler will optimize out all the code of the statement, while still
checking it.
|
|
Checking conditional defines with #ifdef is error prone because we don't
get a compiler warning when the define wrongly is missing.
Instead, always define it to either 0 or 1.
The benefit is also that now we can use NL_DEBUG in C (not only in the
preprocessor).
|
|
We have those defines. Use them throughout.
|
|
For building all of our source files, we should include first
"config.h". This ensures that we have for example NL_DEBUG available.
Actually, "config.h" is included by "nl-default.h". In turn this means,
all our source files should first include "nl-default.h", which drags in
a set of default headers.
Fix the lex/yacc files for that.
|
|
https://github.com/thom311/libnl/pull/373
|
|
https://github.com/thom311/libnl/pull/372
|
|
|
|
|
|
|
|
|
|
https://github.com/thom311/libnl/issues/370
https://github.com/thom311/libnl/pull/371
|
|
|
|
Add headers from 6995e2de6891c724bfeb2db33d7b87775f913ad1 (tag: v6.4).
|
|
Coverity warns about this. Avoid it.
Error: CLANG_WARNING: [#def3]
libnl-3.8.0/lib/addr.c:324:3: warning[deadcode.DeadStores]: Value stored to 'len' is never read
# 322| !strcasecmp(str, "any")) {
# 323|
# 324|-> len = 0;
# 325|
# 326| switch (hint) {
|
|
Coverity doesn't like this. Workaround.
Error: CPPCHECK_WARNING (CWE-190): [#def97]
libnl-3.8.0/lib/route/tc.c:681: error[integerOverflow]: Signed integer overflow for expression '1<<i'.
# 679|
# 680| for (i = 0; i < 32; i++)
# 681|-> if ((1 << i) == cell_size)
# 682| return i;
# 683|
|
|
Don't return -1, that is not an error code.
Also, return early from function.
|
|
Fixes: 49c20efaa783 ('xfrm: fix crashes in case of ENOMEM')
|
|
Coverity really wants to warn if a time_t is cast to 32 bits.
We use time() here to get (some very bad) randomness. The loss
of the upper bits is the least of the problems.
Work around the coverity warning by also the higher bits.
Error: Y2K38_SAFETY (CWE-197): [#def12]
libnl-3.8.0/lib/socket.c:76: store_truncates_time_t: A "time_t" value is stored in an integer with too few bits to accommodate it. The expression "time(NULL)" is cast to "uint32_t".
# 74|
# 75| if (idx_state == 0) {
# 76|-> uint32_t t = (uint32_t) time(NULL);
# 77|
# 78| /* from time to time (on average each 2^15 calls), the idx_state will
Error: Y2K38_SAFETY (CWE-197): [#def13]
libnl-3.8.0/lib/socket.c:193: store_truncates_time_t: A "time_t" value is stored in an integer with too few bits to accommodate it. The expression "time(NULL)" is cast to "unsigned int".
# 191| sk->s_local.nl_family = AF_NETLINK;
# 192| sk->s_peer.nl_family = AF_NETLINK;
# 193|-> sk->s_seq_next = (unsigned int) time(NULL);
# 194| sk->s_seq_expect = sk->s_seq_next;
# 195|
|
|
Update the Fedora container for github. In particular, this updates
the used version of clang-format for formatting the C sources.
|
|
Build a container image for the code formatting, and run that.
|
|
Coverity warns "Signed integer overflow for expression '1<<31'".
|
|
Coverity warns about this. Try to avoid the warning by casting.
We only use time() to get some (very little) entropy. The cast
is fine.
|
|
Not important, but coverity flags this. Avoid it.
|
|
rtnl_act_append() either takes ownership of the argument, or does
nothing (on error). This pattern is hard to get right. In the past,
there were various bugs at this place.
Add two wrappers _rtnl_act_append_get() and _rtnl_act_append_take()
which consistently don't take ownership of the pointer or take it.
Also, in functions like rtnl_flower_append_action() only set the mask
after successfully modifying the data.
|
|
Using rtnl_act_append() correctly is hard.
Fixes: ef46de143206 ('route/cls: add flower classifier')
|
|
Fixes: 105a6be10a5f ('route: use cleanup macro in rtnl_act_parse()')
|
|
|
|
|
|
This is useful for a very special case. For a cleanup macro
that calls rtnl_act_put_all(). In that case, we have a auto
variable `struct rtnl_act *head`, but the cleanup function
expects as argument `&head`.
|
|
Coverity warns:
Error: RESOURCE_LEAK (CWE-772): [#def2]
libnl-3.8.0/lib/route/act.c:421: alloc_fn: Storage is returned from allocation function "rtnl_act_alloc".
libnl-3.8.0/lib/route/act.c:421: var_assign: Assigning: "act" = storage returned from "rtnl_act_alloc()".
libnl-3.8.0/lib/route/act.c:426: var_assign: Assigning: "tc" = "act".
libnl-3.8.0/lib/route/act.c:438: noescape: Resource "tc" is not freed or pointed-to in "rtnl_tc_set_kind".
libnl-3.8.0/lib/route/act.c:455: leaked_storage: Variable "tc" going out of scope leaks the storage it points to.
libnl-3.8.0/lib/route/act.c:455: leaked_storage: Variable "act" going out of scope leaks the storage it points to.
# 453| tc_act_stats_policy);
# 454| if (err < 0)
# 455|-> return err;
# 456|
# 457| if (tb3[TCA_STATS_BASIC]) {
Fixes: 05bd6366387c ('add support for TC action statistics')
|
|
Seems the static array indices can confuse coverity. I think coverity is wrong
here, regardless, change it.
libnl-3.8.0/include/base/nl-base-utils.h:683: overrun-buffer-arg: Overrunning buffer pointed to by "buf" of 16 bytes by passing it to a function which accesses it at byte offset 45 using argument "(addr_family == 2) ? 16 : 46" (which evaluates to 46).
# 681| * and a suitably large buffer, it cannot. Assert for that. */
# 682|
# 683|-> r = (char *)inet_ntop(addr_family, addr, buf,
# 684| (addr_family == AF_INET) ? INET_ADDRSTRLEN :
# 685| INET6_ADDRSTRLEN);
|
|
Adds the rtnl_link_can_get_device_stats function to get xstats of a
CAN-bus link as a can_device_stats struct.
https://github.com/thom311/libnl/pull/368
|
|
This seems good for our test builds.
This can result in additional uninitialized variable warnings about
autovariables with cleanup attribute.
This flag is also enabled by default on Fedora package build ([1]).
[1] https://src.fedoraproject.org/rpms/redhat-rpm-config//blob/rawhide/f/buildflags.md
|
