This is not important for nl_object_identical() which only considers
the required attributes. But for using nl_object_diff() or nl_object_filter(),
all attributes must be compared.
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: Thomas Haller <thaller@redhat.com>
Indicate this change of behavior with the capability
NL_CAPABILITY_ROUTE_ADDR_COMPARE_CACHEINFO.
This is a behavioral change as we now consider an additional
field when comparing addresses.
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: Thomas Haller <thaller@redhat.com>
Older kernels without patch a3d1289126e7b14307074b76bf1677015ea5036f do
not support rtnl_getlink() by ifname. Detect this situation and fail
with -NLE_OPNOTSUPP instead of -NLE_INVAL.
This changes behavior in returning a different error code for this
case.
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: Thomas Haller <thaller@redhat.com>
nl_pickup() converts error codes from netlink into
nl error codes using nl_syserr2nlerr(). The latter function
mangles different error codes to the same nl error code.
Add a new function, that returns both the nl error code
and the original error code.
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: Thomas Haller <thaller@redhat.com>
idiagnl_exts2str() is used to print req->idiag_ext,
which is the extention types combined as flags, such as
( (1 << (INET_DIAG_TOS - 1)) | (1 << (INET_DIAG_MEMINFO - 1)) ).
This function was wrong from the beginning because the string lookup
array had indexes such as 'INET_DIAG_TOS', instead of '(1 << (INET_DIAG_TOS - 1)'.
Fix also idiagnl_attrs2str() which now converts an extension
type (e.g. INET_DIAG_TOS) to a string. Still this function is
deprecated, as the function name is not clear and it is not
used or very useful.
Fixes: 22eb2569a5
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: Thomas Haller <thaller@redhat.com>
Deprecate all of the IDIAG_* defines in 'idiagnl.h'.
Names exposed by libnl should have the prefix IDIAGNL_*
instead.
But actually, most of these values were copies from other
system headers. The user should instead use the original defines
anyway.
Also, several fixes of the idiag_msg_obj, which had a broken
oo_clone() function and missing oo_compare().
http://lists.infradead.org/pipermail/libnl/2014-November/001726.html
Signed-off-by: Thomas Haller <thaller@redhat.com>
Having a oo_keygen() function only makes sense together with a
oo_compare() function because after hashing, you still have to compare
the objects for equality (in case of hash collission).
Fixes: 9c066b9271
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: Thomas Haller <thaller@redhat.com>
A oo_clone() function is only needed for complex sturctures
to deep copy an object.
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: Thomas Haller <thaller@redhat.com>
For one, we did not clone all pointer values. Hence, every cloned
object was very broken and resulted in dangling pointers and
double free/unref.
Apparently nobody was really using this function up to now.
Also, fix the return cases for NLE_NOMEM, so that we did not assume
ownership of pointers in 'src'.
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: Thomas Haller <thaller@redhat.com>
These values mirror TCP_* socket states from 'netinit/tcp.h'.
There is no good reason to expose a copy of those values.
User space should use the original values (if they care).
The only value that is actually useful is IDIAGNL_SS_ALL.
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: Thomas Haller <thaller@redhat.com>
IDIAG_ATTR_* were a copy of the INET_DIAG_* extension kernel
flags. Redefining them is wrong, user space should continue
to use the values provided via the kernel headers.
Also they were misused as change flags (ce_mask), which they are not.
Deprecate the IDIAG_ATTR_* flags and redefine them to what the
originally are: INET_DIAG_*.
Also deprecated idiagnl_attrs2str() because there is already
idiagnl_exts2str(). idiagnl_attrs2str() in the sense of libnl change
flags (ce_mask) makes no sense.
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: Thomas Haller <thaller@redhat.com>
This internal structure should never have been exposed publically.
It as introduced as public API in v3.2.23.
For now, don't remove it as it would be an API/ABI break.
Fixes: 81d2b1d509
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: Thomas Haller <thaller@redhat.com>
Taken from upstream kernel commit bfe01a5ba2490f299e1d2d5508cbbbadd897bbe9
(v3.17), file 'include/uapi/linux/inet_diag.h'.
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: Thomas Haller <thaller@redhat.com>
When oo_clone() fails, the new object is freed. Make sure 'dst'
does not own parts of 'src'.
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: Thomas Haller <thaller@redhat.com>
When we test idiag on 3.4 kernel, we always get ERANGE.
This is because libnl has its own copy for SK_MEMINFO_*,
which is actually newer than 3.4, where SK_MEMINFO_VARS
is larger than kernel's.
We add a copy from latest kernel, so on older kernel
libnl should still compile. Note, for kernel < 3.6
we don't have SK_MEMINFO_BACKLOG, we have to relax
the minlen.
'sock_diag.h' comes from v3.17 kernel sources
(bfe01a5ba2490f299e1d2d5508cbbbadd897bbe9), file
'include/uapi/linux/sock_diag.h'.
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: Thomas Haller <thaller@redhat.com>
Without ->oo_keygen, libnl will use linear search
for cache objects. This is extremely slow for idiag
when we have a lot of TCP connections. Provide a
hash function for idiag so that libnl will be able
to lookup a hashtable.
http://lists.infradead.org/pipermail/libnl/2014-November/001715.html
Cc: Thomas Graf <tgraf@suug.ch>
Cc: Thomas Haller <thaller@redhat.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: Thomas Haller <thaller@redhat.com>
The cache pickup does a check for duplicates for some time now.
However, the pickup in nl_cache_refill() explicitely clears the
cache, so the pickup doesn't need to actually do the dupe check,
as uniqueness is already guaranteed be Netlink subsystem.
And avoiding the dup check is beneficitial for performance
reasons, as the current algorithm is O(n^2).
http://lists.infradead.org/pipermail/libnl/2014-October/001680.html
Fixes: 96bb7c9a4c
Signed-off-by: Holger Eitzenberger <holger@eitzenberger.org>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: Thomas Haller <thaller@redhat.com>
- whitespace
- fix newline and whitespace in output of u32_dump_details()
- let rtnl_u32_del_mark() clear U32_ATTR_MARK
Signed-off-by: Thomas Haller <thaller@redhat.com>
All the io_alloc() implementation unconditionally allocated
new memory, thus leaking memory if called more then once.
Fix io_alloc() implementations not to allocate new memory
if not needed.
This happens for example in link_msg_parser() which first
calls rtnl_link_set_type():
#0 macvlan_alloc (link=0x609d50) at route/link/macvlan.c:56
#1 0x00007ffff7b99a78 in rtnl_link_set_type (link=link@entry=0x609d50, type=type@entry=0x609a94 "macvlan") at route/link.c:2233
#2 0x00007ffff7b99c28 in link_msg_parser (ops=<optimized out>, who=<optimized out>, n=<optimized out>, pp=0x7fffffffd870) at route/link.c:547
#3 0x00007ffff7dea109 in nl_cache_parse (ops=0x7ffff7dd8600 <rtnl_link_ops>, who=0x603338, nlh=0x6098a0, params=0x7fffffffd870) at cache.c:914
#4 0x00007ffff7dea15b in update_msg_parser (msg=<optimized out>, arg=<optimized out>) at cache.c:668
#5 0x00007ffff7def7bf in nl_cb_call (msg=<optimized out>, type=<optimized out>, cb=<optimized out>) at ../include/netlink-private/netlink.h:142
#6 recvmsgs (cb=0x6057a0, sk=0x6034c0) at nl.c:952
#7 nl_recvmsgs_report (sk=sk@entry=0x6034c0, cb=cb@entry=0x6057a0) at nl.c:1003
#8 0x00007ffff7defb79 in nl_recvmsgs (sk=sk@entry=0x6034c0, cb=cb@entry=0x6057a0) at nl.c:1027
#9 0x00007ffff7de9668 in __cache_pickup (sk=0x6034c0, cache=0x603510, param=param@entry=0x7fffffffd870) at cache.c:701
#10 0x00007ffff7dea08d in nl_cache_pickup (sk=<optimized out>, cache=<optimized out>) at cache.c:753
#11 0x0000000000400d56 in main ()
and later ops->io_parse():
#0 macvlan_alloc (link=0x609d50) at route/link/macvlan.c:56
#1 0x00007ffff7baae9d in macvlan_parse (link=0x609d50, data=<optimized out>, xstats=<optimized out>) at route/link/macvlan.c:79
#2 0x00007ffff7b99c80 in link_msg_parser (ops=<optimized out>, who=<optimized out>, n=<optimized out>, pp=0x7fffffffd870) at route/link.c:567
#3 0x00007ffff7dea109 in nl_cache_parse (ops=0x7ffff7dd8600 <rtnl_link_ops>, who=0x603338, nlh=0x6098a0, params=0x7fffffffd870) at cache.c:914
#4 0x00007ffff7dea15b in update_msg_parser (msg=<optimized out>, arg=<optimized out>) at cache.c:668
#5 0x00007ffff7def7bf in nl_cb_call (msg=<optimized out>, type=<optimized out>, cb=<optimized out>) at ../include/netlink-private/netlink.h:142
#6 recvmsgs (cb=0x6057a0, sk=0x6034c0) at nl.c:952
#7 nl_recvmsgs_report (sk=sk@entry=0x6034c0, cb=cb@entry=0x6057a0) at nl.c:1003
#8 0x00007ffff7defb79 in nl_recvmsgs (sk=sk@entry=0x6034c0, cb=cb@entry=0x6057a0) at nl.c:1027
#9 0x00007ffff7de9668 in __cache_pickup (sk=0x6034c0, cache=0x603510, param=param@entry=0x7fffffffd870) at cache.c:701
#10 0x00007ffff7dea08d in nl_cache_pickup (sk=<optimized out>, cache=<optimized out>) at cache.c:753
#11 0x0000000000400d56 in main ()
https://github.com/thom311/libnl/issues/59
Signed-off-by: Thomas Haller <thaller@redhat.com>
libnl3 contains a private copy of some kernel header files.
Normally, users are expected to install libnl and specify
-I$PREFIX/include/libnl3 as include path. As the private kernel
header files are not installed, this works fine.
However, it can be convenient to build against the libnl source
directory, without installing libnl. In this case, the private kernel
header files shaddow the system provided ones. This is undesired.
Move these files to a different directory to avoid this clash.
http://lists.infradead.org/pipermail/libnl/2014-September/001645.html
Signed-off-by: Thomas Haller <thaller@redhat.com>
Acked-by: Thomas Graf <tgraf@suug.ch>
https://github.com/thom311/libnl/pull/62
[thaller@redhat.com: I changed the original patches
to indent with tabs (vs. spaces) and reworded the
commit messages.]
Signed-off-by: Thomas Haller <thaller@redhat.com>
[thaller@redhat.com: I modified the condition "if (data && len)"
in the original patch to just check "len > 0".
Note that all call sites of meta_alloc() make sure to pass a
valid data pointer with a non-zero length (anything else would
be a bug). But indeed, calling memcpy with invalid src pointer
is undefined behavior, even if len is zero.]
Signed-off-by: Thomas Haller <thaller@redhat.com>
