With llvm 7.0 or earlier, the map symbol type is STT_NOTYPE.
-bash-4.4$ cat t.c
__attribute__((section("maps"))) int g;
-bash-4.4$ clang -target bpf -O2 -c t.c
-bash-4.4$ readelf -s t.o
Symbol table '.symtab' contains 2 entries:
Num: Value Size Type Bind Vis Ndx Name
0: 0000000000000000 0 NOTYPE LOCAL DEFAULT UND
1: 0000000000000000 0 NOTYPE GLOBAL DEFAULT 3 g
The following llvm commit enables BPF target to generate
proper symbol type and size.
commit bf6ec206615b9718869d48b4e5400d0c6e3638dd
Author: Yonghong Song <yhs@fb.com>
Date: Wed Sep 19 16:04:13 2018 +0000
[bpf] Symbol sizes and types in object file
Clang-compiled object files currently don't include the symbol sizes and
types. Some tools however need that information. For example, ctfconvert
uses that information to generate FreeBSD's CTF representation from ELF
files.
With this patch, symbol sizes and types are included in object files.
Signed-off-by: Paul Chaignon <paul.chaignon@orange.com>
Reported-by: Yutaro Hayakawa <yhayakawa3720@gmail.com>
Hence, for llvm 8.0.0 (currently trunk), symbol type will be not NOTYPE, but OBJECT.
-bash-4.4$ clang -target bpf -O2 -c t.c
-bash-4.4$ readelf -s t.o
Symbol table '.symtab' contains 3 entries:
Num: Value Size Type Bind Vis Ndx Name
0: 0000000000000000 0 NOTYPE LOCAL DEFAULT UND
1: 0000000000000000 0 FILE LOCAL DEFAULT ABS t.c
2: 0000000000000000 4 OBJECT GLOBAL DEFAULT 3 g
This patch makes sure bpf library accepts both NOTYPE and OBJECT types
of global map symbols.
Signed-off-by: Yonghong Song <yhs@fb.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
This patch adds support for the End.BPF action of the seg6local
lightweight tunnel. Functions from the BPF lightweight tunnel are
re-used in this patch. Example:
$ ip -6 route add fc00::18 encap seg6local action End.BPF endpoint
obj my_bpf.o sec my_func dev eth0
$ ip -6 route show fc00::18
fc00::18 encap seg6local action End.BPF endpoint my_bpf.o:[my_func]
dev eth0 metric 1024 pref medium
v2: - re-use of print_encap_bpf_prog instead of fprintf
- introduction of "endpoint" keyword for more consistency with
others parameters
Signed-off-by: Mathieu Xhonneux <m.xhonneux@gmail.com>
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
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>
Do not bail out when AF_ALG is not supported by the kernel and
only do so when a map is requested in object ns where we're
calculating the hash. Otherwise, the loader can operate just
fine, therefore lets not fail early when it's not needed.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David Ahern <dsahern@gmail.com>
No need to spam the user with this if it can be fixed gracefully
anyway. Therefore, move it under verbose option.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David Ahern <dsahern@gmail.com>
Perf arrays are handled specially by the kernel, don't request
offload even when used by an offloaded program.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David Ahern <dsahern@gmail.com>
In theory, the path for BPF could exceed the 4K PATH_MAX.
In practice, not really possible. But shut up gcc.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Use strlcpy to avoid cases where sizeof(buf) == strlen(buf)
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
It's useful to be able to tell which section is being processed in the
ELF when this error is triggered, so print that detail.
Signed-off-by: Joe Stringer <joe@wand.net.nz>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
When program is loaded with a specified ifindex, use that
ifindex also when creating maps.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David Ahern <dsahern@gmail.com>
For BPF offload we need to specify the ifindex when program is
loaded now. Extend the bpf common code to accommodate that.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Expose bpf_parse_common() and bpf_load_common() functions
for those users who may want to modify the parameters to
load after parsing is done.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
bpf_parse_common() parses and loads the program. Rename it
accordingly.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Parsing command line is currently done together with potentially
loading a new eBPF program. This makes it more difficult to
provide additional parameters for loading (which may come after
the eBPF program info on the command line).
Split the two (only internally for now). Verbose parameter
has to be saved in struct bpf_cfg_in to be carried between
the stages.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
struct bpf_cfg_in already carries a pointer to sock_filter ops.
It's currently set to a local variable in bpf_parse_opt_tbl(),
shared between parsing and loading stages. Move the array
entirely to struct bpf_cfg_in, this will allow us to split
parsing and loading.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
bpf_parse() will parse command line arguments to find out the
program mode. This mode will later be needed at loading time.
Instead of keeping it locally add it to struct bpf_cfg_in,
this will allow splitting parsing and loading stages.
enum bpf_mode has to be moved to the header file, because C
doesn't allow forward declaration of enums.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Program type is needed both for parsing and loading of
the program. Parsing may also induce the type based on
signatures from __bpf_prog_meta. Instead of passing
the type around keep it in struct bpf_cfg_in.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
If program loading fails before verifier prints its first
message, the verifier log will not be initialized. Always
set the first character of the log buffer to zero to make
sure we don't dump non-printable characters to the terminal.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Consolidate dump of prog info to use bpf_dump_prog_info() when possible.
Moving forward, we want to have a consistent output for BPF progs when
being dumped. E.g. in cls/act case we used to dump tag as a separate
netlink attribute before we had BPF_OBJ_GET_INFO_BY_FD bpf(2) command.
Move dumping tag into bpf_dump_prog_info() as well, and only dump the
netlink attribute for older kernels. Also, reuse bpf_dump_prog_info()
for XDP case, so we can dump tag and whether program was jited, which
we currently don't show.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Just minor nits, e.g. no need to fflush() and instead of returning
right away, just break and close the fd.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
The signedness of char type is implementation dependent, and there are
architectures on which it is unsigned by default. In that case, the
check whether fgetc() returned EOF failed because the return value was
assigned an (unsigned) char variable prior to comparison with EOF (which
is defined to -1). Fix this by using int as type for 'c' variable, which
also matches the declaration of fgetc().
While being at it, fix the parser logic to correctly handle multiple
empty lines and consecutive whitespace and tab characters to further
improve the parser's robustness. Note that this will still detect double
separator characters, so doesn't soften up the parser too much.
Fixes: 3da3ebfca8 ("bpf: Make bytecode-file reading a little more robust")
Cc: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Phil Sutter <phil@nwl.cc>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
This is merely to silence the compiler warning. If write to stderr
failed, assume that printing an error message will fail as well so don't
even try.
Signed-off-by: Phil Sutter <phil@nwl.cc>
If fopen() succeeded but len != PATH_MAX, the function leaks the open
FILE pointer. Fix this by checking len value before calling fopen().
Signed-off-by: Phil Sutter <phil@nwl.cc>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
bpf_parse_string() will now correctly handle:
- Extraneous whitespace,
- OPs on multiple lines and
- overlong file names.
The added feature of allowing to have OPs on multiple lines (like e.g.
tcpdump prints them) is rather a side effect of fixing detection of
malformed bytecode files having random content on a second line, like
e.g.:
| 4,40 0 0 12,21 0 1 2048,6 0 0 262144,6 0 0 0
| foobar
Cc: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Phil Sutter <phil@nwl.cc>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
When bpf fs mount path is from env, behavior is currently broken as
we continue to search in default paths, thus fix this up.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Currently, it's still quite hard to figure out if a prog passed the
verifier, but later gets rejected due to different tail call ownership.
Figure out whether that is the case and provide appropriate error
messages to the user.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Make use of TCA_BPF_ID/TCA_ACT_BPF_ID that we exposed and print the ID
of the programs loaded and use the new BPF_OBJ_GET_INFO_BY_FD command
for dumping further information about the program, currently whether
the attached program is jited.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Add support for map in map in the loader and add a small example program.
The outer map uses inner_id to reference a bpf_elf_map with a given ID
as the inner type. Loading maps is done in three passes, i) all non-map
in map maps are loaded, ii) all map in map maps are loaded based on the
inner_id map spec of a non-map in map with corresponding id, and iii)
related inner maps are attached to the map in map with given inner_idx
key. Pinned objetcs are assumed to be managed externally, so they are
only retrieved from BPF fs.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
When LLVM wrongly generates a rodata relo entry (llvm BZ #33599),
then just bail out instead of probing for prog w/o reloc, which
will fail in this case anyway.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
I noticed we currently don't dump an error message when a pinned
program couldn't be retrieved, thus add a hint to the user.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Jan-Erik reported an assertion in bpf_prog_to_subdir() failed where
type was BPF_PROG_TYPE_UNSPEC, which is only used in bpf_init_env()
to auto-mount and cache the bpf fs mount point.
Therefore, make sure when bpf_init_env() is called multiple times
(f.e. eBPF classifier with eBPF action attached) and bpf_mnt_cached
is set already that the type is also valid. In bpf_init_env(), we're
only interested in the mount point and not a type-specific subdir.
Fixes: e42256699c ("bpf: make tc's bpf loader generic and move into lib")
Reported-by: Jan-Erik Rediger <janerik@rediger.net>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Adds support to configure BPF programs as nexthop actions via the LWT
framework.
Example:
ip route add 192.168.253.2/32 \
encap bpf out obj lwt_len_hist_kern.o section len_hist \
dev veth0
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Now that we made the BPF loader generic as a library, reuse it
for loading XDP programs as well. This basically adds a minimal
start of a facility for iproute2 to load XDP programs. There
currently only exists the xdp1_user.c sample code in the kernel
tree that sets up netlink directly and an iovisor/bcc front-end.
Since we have all the necessary infrastructure in place already
from tc side, we can just reuse its loader back-end and thus
facilitate migration and usability among the two for people
familiar with tc/bpf already. Sharing maps, performing tail calls,
etc works the same way as with tc. Naturally, once kernel
configuration API evolves, we will extend new features for XDP
here as well, resp. extend dumping of related netlink attributes.
Minimal example:
clang -target bpf -O2 -Wall -c prog.c -o prog.o
ip [-force] link set dev em1 xdp obj prog.o # attaching
ip [-d] link # dumping
ip link set dev em1 xdp off # detaching
For the dump, intention is that in the first line for each ip
link entry, we'll see "xdp" to indicate that this device has an
XDP program attached. Once we dump some more useful information
via netlink (digest, etc), idea is that 'ip -d link' will then
display additional relevant program information below the "link/
ether [...]" output line for such devices, for example.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Kernel commit 21116b7068b9 ("bpf: add owner_prog_type and accounted mem
to array map's fdinfo") added support for telling the owner prog type in
case of prog arrays. Give a notification to the user when they differ,
and the program eventually fails to load.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
The log buffer is automatically grown when the verifier output does not
fit into the default buffer size. The number of growing attempts was
not sufficient to reach the maximum buffer size so far.
Perform 9 iterations to reach max and let the 10th one fail.
j:0 i:65536 max:16777215
j:1 i:131072 max:16777215
j:2 i:262144 max:16777215
j:3 i:524288 max:16777215
j:4 i:1048576 max:16777215
j:5 i:2097152 max:16777215
j:6 i:4194304 max:16777215
j:7 i:8388608 max:16777215
j:8 i:16777216 max:16777215
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
This work moves the bpf loader into the iproute2 library and reworks
the tc specific parts into generic code. It's useful as we can then
more easily support new program types by just having the same ELF
loader backend. Joint work with Thomas Graf. I hacked a rough start
of a test suite to make sure nothing breaks [1] and looks all good.
[1] https://github.com/borkmann/clsact/blob/master/test_bpf.sh
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Thomas Graf <tgraf@suug.ch>
