summaryrefslogtreecommitdiffstats
Commit message (Collapse)AuthorAgeFilesLines
...
* tests: silently ignore EACCES for setting uid_map for test namespaceThomas Haller2022-04-221-1/+9
| | | | | | | Seems this can happen, but we probably can just continue with the unit test. Just ignore the error. https://github.com/thom311/libnl/issues/308
* tests: cleanup unshare_user() and use _nltst_fclose()Thomas Haller2022-04-212-9/+20
|
* tests: add _assert_nltst_netns() helperThomas Haller2022-04-211-2/+10
|
* github: test out-of-tree build and "--disable-static"Thomas Haller2022-04-151-2/+29
|
* github: build documentation in CI testThomas Haller2022-04-152-2/+23
|
* build: avoid building check-direct with --disable-staticThomas Haller2022-04-152-0/+4
| | | | | | | | | | | "check-direct" needs to statically link with the libraries, because it wants to test internal ABI, which is hidden in the share libraries. When configuring with "--disable-static", static libs are not build and the test tool cannot be build. Just skip the test in that case. https://github.com/thom311/libnl/issues/306
* tools: fix aborting on failure in "tools/build_release.sh" scriptThomas Haller2022-04-151-2/+2
|
* doc: fix markup error in "doc/route.txt"Thomas Haller2022-04-151-2/+2
| | | | Fixes: d9dc6c20a360 ('ip6vti: Add IPv6 VTI support')
* doc: fix python2-ism in "doc/resolve-asciidoc-refs.py"Thomas Haller2022-04-151-1/+3
|
* libnl-3.6.0 releaselibnl3_6_0Thomas Haller2022-04-144-3/+9
|
* route/mdb: merge branch 'troglobit:mdb-dump-fixes'Thomas Haller2022-04-013-3/+27
|\ | | | | | | https://github.com/thom311/libnl/pull/305
| * route/mdb: add support for MAC multicast entriesJoachim Wiberg2022-03-302-0/+6
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | The kernel bridge now support (permanent) forwarding of MAC multicast using the MDB. Internally the kernel use AF_UNSPEC, but we remap this here to AF_LLC for the benefit for nl_addrs. To test, put `nl-monitor mdb` in the background. Then, with a bridge and at least one port, run the following command: # nl-monitor mdb & # bridge mdb add dev br0 port eth0 grp 01:02:03:c0:ff:ee vid 1 permanent dev 9 port 3 vid 1 proto 0x0000 address 01:02:03:c0:ff:ee Signed-off-by: Joachim Wiberg <troglobit@gmail.com>
| * route/mdb: add missing detils and stats dump callbacksJoachim Wiberg2022-03-301-2/+14
| | | | | | | | | | | | | | | | When using, e.g., nl-monitor to debug the bridge mdb the nl-monitor tool did not dump anything. This change adds the missing stats dump callback to rectify this issue, and also the details callback for completeness. Signed-off-by: Joachim Wiberg <troglobit@gmail.com>
| * nl-monitor: support for setting libnl debug levelJoachim Wiberg2022-03-301-1/+6
| | | | | | | | | | | | | | When debugging subystemns in libnl it's rather handy to use nl-monitor. This change adds support for setting the nl_debug level of libnl. Signed-off-by: Joachim Wiberg <troglobit@gmail.com>
| * nl-monitor: add missing --help to long_opts[]Joachim Wiberg2022-03-301-0/+1
|/ | | | Signed-off-by: Joachim Wiberg <troglobit@gmail.com>
* Check validation type against end of enumjauge-technica2022-03-281-1/+1
| | | | | | | | The enum macsec_validation_type in the Linux Kernel has values 0-2. With the existing check >1, value STRICT (2) cannot be set. The check should be done against the end marker of the enum instead. https://github.com/thom311/libnl/pull/304
* route/link: add VLAN bridge binding flagJoachim Wiberg2022-03-282-4/+6
| | | | | | | | | | | | Adds support for the new VLAN_FLAG_BRIDGE_BINDING, for VLAN interfaces created on top of a VLAN aware bridge. For details, see the kernel patch: https://lore.kernel.org/netdev/20190418173535.22925-1-mmanning@vyatta.att-mail.com/ Signed-off-by: Joachim Wiberg <troglobit@gmail.com> https://github.com/thom311/libnl/pull/303
* github: build unit tests also with "clang"Thomas Haller2022-03-161-0/+6
| | | | In addition to gcc.
* route: assert that "rtnl_link_info_ops" refcount does not drop below zeroThomas Haller2022-03-161-0/+2
|
* lib: merge branch 'th/object-clone-fixes'Thomas Haller2022-03-1626-107/+271
|\ | | | | | | https://github.com/thom311/libnl/pull/301
| * lib: make nl_object_clone() out-of-memory safeThomas Haller2022-03-1620-54/+149
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | nl_object_clone() first does a shallow copy using memcpy(). That is useful, because it can correctly copy simple fields (like numbers). For pointer values, we need to implement oo_clone() to fixup the pointers and get the deep-copy correct. Now, oo_clone() must always follow through, to un-alias the copied pointer. In particular also in the error case. The oo_clone() implementations sometimes fail (with ENOMEM) and just return. In those cases, we must make sure that we don't leave the wrong pointers there. The pointers must be cleared first. Otherwise, any failure (which basically are ENOMEM) leave the object in an inconsistent state, and we double-free/use-after-free the pointers.
| * route: fix ref counting for l_info_ops and io_clone()Thomas Haller2022-03-161-4/+10
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | nl_object_clone() first does a shallow copy, before calling oo_clone() (link_clone()). That means, the pointer values of the link object in link_clone() are invalid (as they alias the pointers from the source object). We need to get the ref-counting for dst->l_info_ops right. It was not. For example, previously when we called io_clone() handler, dst->l_info_ops would still point to the one from src->l_info_ops, but without owning the additional reference. Then we call io_clone(), for example can_clone() for can devices. That one calls first rtnl_link_set_type(), which first calls release_link_info() -- and unrefs the ops, without having owned a reference. Fix that, by getting the reference counting right, before calling io_clone(). Arguably, we now do duplicate work. First taking a ref, then calling rtnl_link_set_type() which releases and retakes the ref. But at least, this way it's correct. This probably did not cause issues before, because the entire ref-counting is mostly useless anyway. It's only used for asserting during rtnl_link_unregister_info() -- and then it checks that the ref count is not positive (but we release too many references, not too few). Anyway. *sigh*.
| * route: drop unnecessary oo_clone() implementation from netconfThomas Haller2022-03-161-11/+0
| | | | | | | | | | | | If the object has no complex data (pointers!), then the base implementation using memcpy() is enough. No need to implement oo_clone().
| * netfilter: make log-msg,queue-msg setters robust against ENOMEMThomas Haller2022-03-162-22/+52
| | | | | | | | | | | | If an error happens (ENOMEM), we should leave the object in a consistent state (e.g. setting log_msg_payload *and* log_msg_payload_len). Or even better, don't modify it at all in the error case.
| * xfrm/sa: clone user_offload in xfrm_sa_clone()Thomas Haller2022-03-161-0/+6
| |
| * xfrm/sa: style cleanup xfrm_sa_clone()Thomas Haller2022-03-161-14/+7
| |
| * utils: add internal _nl_memdup() helperThomas Haller2022-03-161-0/+20
| |
| * lib: add rtnl_link_info_ops_get() and take lock for rtnl_link_info_ops's ↵Thomas Haller2022-03-163-2/+26
| | | | | | | | io_refcnt
| * lib: include <netlink-private/utils.h> in <netlink-private/netlink.h>Thomas Haller2022-03-161-0/+1
|/ | | | | | We have a base-set of tools, under netlink-private. It should not be necessary to include all the bits individually. Just drag this all in. Only downside is that the compiler has more to parse.
* tests: merge branch 'th/tests-netns'Thomas Haller2022-03-1615-48/+627
|\ | | | | | | https://github.com/thom311/libnl/pull/302
| * tests: add unit test for nl_object_clone() and nl_object_diff()Thomas Haller2022-03-163-2/+214
| |
| * tests: add new "netns" test suiteThomas Haller2022-03-164-5/+25
| | | | | | | | | | This suite has fixture/teardown which creates a new netns for the test. No tests implemented yet.
| * tests: add fixture/teardown for tests to run in separate netnsThomas Haller2022-03-162-0/+147
| | | | | | | | | | | | | | | | | | | | | | | | | | We run the unit tests as non-root user, so usually we wouldn't have CAP_NET_ADMIN permissions to change networking. Also, we wouldn't want that our unit tests depend on the networking of the test host (or changes it). For each test, enter a new network namespace (and user/mnt namespace). There we will have the necessary permissions, and we are in full control of the things in the namespace. Note yet used.
| * tests: cleanup creating test suitesThomas Haller2022-03-164-20/+20
| |
| * tests: refactor tests and add n-test-util helper libraryThomas Haller2022-03-169-23/+59
| | | | | | | | | | | | | | | | | | | | | | | | | | | | - we have "check-all.c" and "check-direct.c", which contains the main functions of the actual tests. On the other hand, the other "check-{addr,attr,ematch-tree-clone}.c" files only contained the test suites for "check-all.c". Rename the latter to have a separate name prefix. - rename "tests/util.h" to "tests/cksuite-all.h". It's really the header that declares all the suites. - add a "tests/nl-test-util.c" as a static helper library with test code.
| * netlink: add _NL_N_ELEMENTS() macroThomas Haller2022-03-161-0/+4
| |
| * netlink: add _nl_streq()/_nl_streq0() helperThomas Haller2022-03-161-0/+12
| |
| * netlink: add _nl_auto_nl_socket cleanup macroThomas Haller2022-03-161-0/+5
| |
| * lib: add _nl_close() helperThomas Haller2022-03-161-0/+12
| | | | | | | | | | | | Closing an invalid filedescriptor is a serious bug, because we need to know whether an integer at hand is valid or not (cannot close arbitrary files). Add _nl_close() wrapper, which asserts against EBADF.
| * clang-format: add ".clang-format" from linux kernelThomas Haller2022-03-161-0/+131
| | | | | | | | | | | | | | | | | | | | | | | | Libnl historically is closely related to kernel development and follows the same coding style. Copy kernel's format specification, in the hope to be useful. There is no plan to reformat existing code, but with this file in place, a developer can choose to selectively reformat the code. Taken from linux kernel at [1]. The file is unmodified, except for the "ForEachMacros". [1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/.clang-format?id=56e337f2cf1326323844927a04e9dbce9a244835
| * github: build tests with "-std=gnu11"Thomas Haller2022-03-161-1/+1
|/ | | | | | | | | Libnl3 is C11. Not sure this was defined somewhere, but at this point, it seems a safe requirement to make. Also, we make use of gcc-isms (typeof(), attribute(cleanup()), expression statements). Both gcc and clang support that. These C extensions are too good to miss.
* github: split tests in separate stepsThomas Haller2022-03-161-3/+9
|
* build: add "check-progs" make target to build unit testsThomas Haller2022-03-161-0/+8
|
* route/cls: add TCA_FLOWER_KEY_VLAN_ETH_TYPE to "flower_policy" policyThomas Haller2022-03-151-10/+11
|
* route/cls: return -NLE_INVAL in case rtnl_tc_data_peek() failsThomas Haller2022-03-151-6/+6
| | | | | | The problem here is not really NOMEM (because rtnl_tc_data_peek() wouldn't allocate new memory. The problem is, that the cls instance has no such data field. Return -NLE_INVAL instead.
* route/cls: merge branch 'westermo:cls-flower'Thomas Haller2022-03-155-0/+804
|\ | | | | | | https://github.com/thom311/libnl/pull/200
| * route/cls: no need to copy simple fields in flower_clone()Thomas Haller2022-03-151-12/+0
| | | | | | | | | | | | nl_object_clone() already does a shallow clone using memcpy(). So all simple fields are already initialized. We only need the oo_clone() implementation to get the deep-copy right.
| * route/cls: make output pointers in rtnl_flower_get_{src,dst}_mac() optionalThomas Haller2022-03-151-5/+7
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | We often require that a valid output pointer is passed to the getters. But here, let's be forgiving. Being more forgiving may be inconsistent, but on the safe side: the user is still welcome to assume they must provide a valid output pointer. Also, we always need to initialize the output mask, not only with `f->cf_mask & FLOWER_ATTR_DST_MAC_MASK`. The reason is that the caller cannot know whether the mask is present, so conditionally initializing the output is error prone (it requires the caller to NUL initialize first). Also, "f->cf_dst_mac_mask" really should be zero initialized, if the mask indicates that it's unset. This means, we can just always memcpy the mask.
| * route/cls: adjust whitspace/indentationThomas Haller2022-03-152-162/+162
| |
| * route/cls: use SPDX-License-IdentifierThomas Haller2022-03-152-14/+2
| |
generated by cgit v0.12 at 2025-04-06 15:43:20 (GMT)