Implement loading of .BTF section from object file and build up
internal table for retrieving key/value id related to maps in
the BPF program. Latter is done by setting up struct btf_type
table.
One of the issues is that there's a disconnect between the data
types used in the map and struct bpf_elf_map, meaning the underlying
types are unknown from the map description. One way to overcome
this is to add a annotation such that the loader will recognize
the relation to both. BPF_ANNOTATE_KV_PAIR(map_foo, struct key,
struct val); has been added to the API that programs can use.
The loader will then pick the corresponding key/value type ids and
attach it to the maps for creation. This can later on be dumped via
bpftool for introspection.
Example with test_xdp_noinline.o from kernel selftests:
[...]
struct ctl_value {
union {
__u64 value;
__u32 ifindex;
__u8 mac[6];
};
};
struct bpf_map_def __attribute__ ((section("maps"), used)) ctl_array = {
.type = BPF_MAP_TYPE_ARRAY,
.key_size = sizeof(__u32),
.value_size = sizeof(struct ctl_value),
.max_entries = 16,
.map_flags = 0,
};
BPF_ANNOTATE_KV_PAIR(ctl_array, __u32, struct ctl_value);
[...]
Above could also further be wrapped in a macro. Compiling through LLVM and
converting to BTF:
# llc --version
LLVM (http://llvm.org/):
LLVM version 7.0.0svn
Optimized build.
Default target: x86_64-unknown-linux-gnu
Host CPU: skylake
Registered Targets:
bpf - BPF (host endian)
bpfeb - BPF (big endian)
bpfel - BPF (little endian)
[...]
# clang [...] -O2 -target bpf -g -emit-llvm -c test_xdp_noinline.c -o - |
llc -march=bpf -mcpu=probe -mattr=dwarfris -filetype=obj -o test_xdp_noinline.o
# pahole -J test_xdp_noinline.o
Checking pahole dump of BPF object file:
# file test_xdp_noinline.o
test_xdp_noinline.o: ELF 64-bit LSB relocatable, *unknown arch 0xf7* version 1 (SYSV), with debug_info, not stripped
# pahole test_xdp_noinline.o
[...]
struct ctl_value {
union {
__u64 value; /* 0 8 */
__u32 ifindex; /* 0 4 */
__u8 mac[0]; /* 0 0 */
}; /* 0 8 */
/* size: 8, cachelines: 1, members: 1 */
/* last cacheline: 8 bytes */
};
Now loading into kernel and dumping the map via bpftool:
# ip -force link set dev lo xdp obj test_xdp_noinline.o sec xdp-test
# ip a
1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 xdpgeneric/id:227 qdisc noqueue state UNKNOWN group default qlen 1000
link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
inet 127.0.0.1/8 scope host lo
valid_lft forever preferred_lft forever
inet6 ::1/128 scope host
valid_lft forever preferred_lft forever
[...]
# bpftool prog show id 227
227: xdp tag a85e060c275c5616 gpl
loaded_at 2018-07-17T14:41:29+0000 uid 0
xlated 8152B not jited memlock 12288B map_ids 381,385,386,382,384,383
# bpftool map dump id 386
[{
"key": 0,
"value": {
"": {
"value": 0,
"ifindex": 0,
"mac": []
}
}
},{
"key": 1,
"value": {
"": {
"value": 0,
"ifindex": 0,
"mac": []
}
}
},{
[...]
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David Ahern <dsahern@gmail.com>
Import btf.h from kernel headers at commit
2aa4a3378ad0 ("Merge git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next")
which is the last sync point.
Signed-off-by: David Ahern <dsahern@gmail.com>
Currently NETNS_RUN_DIR is hardcoded and refers to /var/run/netns.
However, some systems (e.g. Android) doesn't have /var
which results in error attempts to create network namespaces on these
systems. This change makes NETNS_RUN_DIR configurable at build time
by allowing to pass environment variable to make command.
Also, this change makes /etc/netns directory configurable through
NETNS_ETC_DIR environment variable.
For example: ./configure && NETNS_RUN_DIR=/mnt/vendor/netns make
Tested: verified that iproute2 with configuration mentioned above
creates namespaces in /mnt/vendor/netns
Signed-off-by: Pavel Maltsev <pavelm@google.com>
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
print_uint() will silently promote its variable type to uint64_t, but there
is nothing that ensures that the format string specifier passed along with
it fits (and the function name suggest to pass "%u").
Fix this by changing print_uint() to use a native 'unsigned int' type, and
introduce a separate print_u64() function for printing 64-bit values. All
call sites that were actually printing 64-bit values using print_uint() are
converted to use print_u64() instead.
Since print_int() was already using native int types, just add a
print_s64() to match, but don't convert any call sites. For symmetry,
also add a print_luint() method (with no users).
Signed-off-by: Toke Høiland-Jørgensen <toke@toke.dk>
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Part of upstream commit
4bbb3e0e8239 ("net: Fix vlan untag for bridge and vlan_dev with reorder_hdr off")
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
ip vrf exec requires root or CAP_NET_ADMIN, CAP_SYS_ADMIN and
CAP_DAC_OVERRIDE. It is not possible to run unprivileged commands like
ping as non-root or non-cap-enabled due to this requirement.
To allow users and administrators to safely add the required
capabilities to the binary, drop all capabilities on start if not
invoked with "vrf exec".
Update the manpage with the requirements.
Signed-off-by: Luca Boccassi <bluca@debian.org>
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
In iproute2 package, the updates of UAPIs files are performed
after the needed feature lands in kernel's net-next tree.
Such development flow created delays to the rdma tool developers,
who uses rdma-next tree as a basis for their work.
Move RDMA UAPI file to be under rdma/ folder, so whole responsibility
of syncing this file will be on them.
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Reviewed-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: David Ahern <dsahern@gmail.com>
Conflicts:
bridge/mdb.c
Updated bridge/bridge.c per removal of check_if_color_enabled by commit
1ca4341d2c ("color: disable color when json output is requested")
Signed-off-by: David Ahern <dsahern@gmail.com>
There is a couple of places where we report error in case of no network
device is found. In all of them we output message in the same format to
stderr and either return -1 or 1 to the caller or exit with -1.
Introduce new helper function nodev() that takes name of the network
device caused error and returns -1 to it's caller. Either call exit()
or return to the caller to preserve behaviour before change.
Use -nodev() in traffic control (tc) code to return 1.
Simplify expression for checking for argument being 0/NULL in @if
statement.
Signed-off-by: Serhey Popovych <serhe.popovych@gmail.com>
Add support for devlink resource abstraction. The resources are
represented by a tree based structure and are identified by a name and
a size. Some resources can present their real time occupancy.
First the resources exposed by the driver can be observed, for example:
$devlink resource show pci/0000:03:00.0
pci/0000:03:00.0:
name kvd size 245760 unit entry
resources:
name linear size 98304 occ 0 unit entry size_min 0 size_max 147456 size_gran 128
name hash_double size 60416 unit entry size_min 32768 size_max 180224 size_gran 128
name hash_single size 87040 unit entry size_min 65536 size_max 212992 size_gran 128
Some resource's size can be changed. Examples:
$devlink resource set pci/0000:03:00.0 path /kvd/hash_single size 73088
$devlink resource set pci/0000:03:00.0 path /kvd/hash_double size 74368
The changes do not apply immediately, this can be validate by the 'size_new'
attribute, which represents the pending changed size. For example
$devlink resource show pci/0000:03:00.0
pci/0000:03:00.0:
name kvd size 245760 unit entry size_valid false
resources:
name linear size 98304 size_new 147456 occ 0 unit entry size_min 0 size_max 147456 size_gran 128
name hash_double size 60416 unit entry size_min 32768 size_max 180224 size_gran 128
name hash_single size 87040 unit entry size_min 65536 size_max 212992 size_gran 128
In case of a pending change the nested resources present an indication
for a valid configuration of its children (sum of its children sizes
doesn't exceed the parent's size).
In order for the changes to take place hot reload is needed. The hot
reload through devlink will be introduced in the following patch.
Signed-off-by: Arkadi Sharshevsky <arkadis@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Instead of declaring -color and -json exclusive, ignore -color when
-json is provided. The rationale is to allow to put -color in an alias
for ip while still being able to use -json. -color is merely a
presentation suggestion and we can assume there is nothing to color in
the JSON output.
Signed-off-by: Vincent Bernat <vincent@bernat.im>
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Even in brief mode the interface name should be printed
in color if desired. This makes output consistent across
regular and brief mode.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David Ahern <dsahern@gmail.com>
There is at least three places implementing same things: two in
ipaddress.c print_linkinfo() & print_linkinfo_brief() and one in
bridge/link.c.
They are diverge from each other very little: bridge/link.c does not
support JSON output at the moment and print_linkinfo_brief() does not
handle IFLA_LINK_NETNS case.
Introduce and use print_name_and_link() routine to handle name@link
output in all possible variations; respect IFLA_LINK_NETNS attribute to
handle case when link is in different namespace; use ll_idx_n2a() for
interface name instead of "<nil>" to share logic with other code (e.g.
ll_name_to_index() and ll_index_to_name()) supporting such template.
Signed-off-by: Serhey Popovych <serhe.popovych@gmail.com>
Signed-off-by: David Ahern <dsahern@gmail.com>
Be consistent in handling of IFLA_IFNAME attribute in all places: if
there is no attribute report bug to stderr and use ll_idx_n2a() as
last measure to get name in "if%u" format instead of "<nil>".
Use check_ifname() to validate network device name: this catches both
unexpected return from kernel and ll_idx_n2a().
Signed-off-by: Serhey Popovych <serhe.popovych@gmail.com>
Signed-off-by: David Ahern <dsahern@gmail.com>
Now all users of ll_idx_n2a() replaced with ll_index_to_name() we can
move it's functionality to ll_index_to_name() and implement index to
name conversion using snprintf() and "if%u".
Use %u specifier in "if%..." template consistently: network device
indexes are always greather than zero.
Also introduce ll_idx_n2a() conterpart: ll_idx_a2n() that is used
to translate name of the "if%u" form to index using sscanf().
Signed-off-by: Serhey Popovych <serhe.popovych@gmail.com>
Signed-off-by: David Ahern <dsahern@gmail.com>
After commit a233caa0aa ("json: make pretty printing optional") I get
following build failure:
LINK rtmon
../lib/libutil.a(json_print.o): In function `new_json_obj':
json_print.c:(.text+0x35): undefined reference to `show_pretty'
collect2: error: ld returned 1 exit status
make[1]: *** [rtmon] Error 1
make: *** [all] Error 2
It is caused by missing show_pretty variable in rtmon.
On the other hand tc/tc.c there are two distinct variables and single
matches() call that handles -pretty option thus setting show_pretty
will never happen. Note that since commit 44dcfe8201 ("Change
formatting of u32 back to default") show_pretty is used in tc/f_u32.c
so this is first place where -pretty introduced.
Furthermore other utilities like misc/ifstat.c and misc/nstat.c define
pretty variable, however only for their own purposes. They both support
JSON output and thus depend show_pretty in new_json_obj().
Assuming above use common variable to represent -pretty option, define
it in utils.c and declare in utils.h that is commonly used. Replace
show_pretty with pretty.
Fixes: a233caa0aa ("json: make pretty printing optional")
Signed-off-by: Serhey Popovych <serhe.popovych@gmail.com>
Signed-off-by: David Ahern <dsahern@gmail.com>
Initializing @inet_prefix using C initializers or memset() seems
inefficient and unnecessary: only small part of ->data[] field will be
used to store address corresponding to ->family.
Instead initialize ->flags with zero and assume no other fields accessed
before checking corresponding bits in ->flags. For example special
helpers (e.g. is_addrtype_*()) can be used to ensure that @inet_prefix
contains valid ip or ipv6 address.
Signed-off-by: Serhey Popovych <serhe.popovych@gmail.com>
Signed-off-by: David Ahern <dsahern@gmail.com>
Add JSON and color output formatting to ip route command.
Similar to existing address and link output.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David Ahern <dsahern@gmail.com>
Since JSON is intended for programmatic consumption, it makes
sense for the default output format to be concise as possible.
For programmer and other uses, it is helpful to keep the pretty
whitespace format; therefore enable it with -p flag.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David Ahern <dsahern@gmail.com>
David Ahern
Daniel Borkmann
Stephen Hemminger
Pavel Maltsev
Toke Høiland-Jørgensen
Steve Wise
Luca Boccassi
Leon Romanovsky
Serhey Popovych
Arkadi Sharshevsky
Vincent Bernat
Stephen Hemminger