1libxdp(3) libxdp - library for loading XDP programs libxdp(3)
2
6 This directory contains the files for the libxdp library for attaching
7 XDP programs to network interfaces and using AF_XDP sockets. The li‐
8 brary is fairly lightweight and relies on libbpf to do the heavy lift‐
9 ing for processing eBPF object files etc.
10 Libxdp provides two primary features on top of libbpf. The first is the
13 ability to load multiple XDP programs in sequence on a single network
14 device (which is not natively supported by the kernel). This support
15 relies on the freplace functionality in the kernel, which makes it pos‐
16 sible to attach an eBPF program as a replacement for a global function
17 in another (already loaded) eBPF program. The second main feature is
18 helper functions for configuring AF_XDP sockets as well as reading and
19 writing packets from these sockets.
20 Some of the functionality provided by libxdp depends on particular ker‐
23 nel features; see the "Kernel feature compatibility" section below for
24 details.
25 Using libxdp from an application
28 Basic usage of libxdp from an application is quite straight forward.
29 The following example loads, then unloads, an XDP program from the 'lo'
30 interface:
31 #define IFINDEX 1
33 struct xdp_program *prog;
35 int err;
36 prog = xdp_program__open_file("my-program.o", "section_name", NULL);
38 err = xdp_program__attach(prog, IFINDEX, XDP_MODE_NATIVE, 0);
39 if (!err)
41 xdp_program__detach(prog, IFINDEX, XDP_MODE_NATIVE, 0);
42 xdp_program__close(prog);
44 The xdp_program structure is an opaque structure that represents a sin‐
47 gle XDP program. libxdp contains functions to create such a struct ei‐
48 ther from a BPF object file on disk, from a libbpf BPF object, or from
49 an identifier of a program that is already loaded into the kernel:
50 struct xdp_program *xdp_program__from_bpf_obj(struct bpf_object *obj,
52 const char *section_name);
53 struct xdp_program *xdp_program__find_file(const char *filename,
54 const char *section_name,
55 struct bpf_object_open_opts *opts);
56 struct xdp_program *xdp_program__open_file(const char *filename,
57 const char *section_name,
58 struct bpf_object_open_opts *opts);
59 struct xdp_program *xdp_program__from_fd(int fd);
60 struct xdp_program *xdp_program__from_id(__u32 prog_id);
61 struct xdp_program *xdp_program__from_pin(const char *pin_path);
62 The functions that open a BPF object or file need the function name of
65 the XDP program as well as the file name or object, since an ELF file
66 can contain multiple XDP programs. The xdp_program__find_file() func‐
67 tion takes a filename without a path, and will look for the object in
68 LIBXDP_OBJECT_PATH which defaults to /usr/lib/bpf (or /usr/lib64/bpf on
69 systems using a split library path). This is convenient for applica‐
70 tions shipping pre-compiled eBPF object files.
71 The xdp_program__attach() function will attach the program to an inter‐
74 face, building a dispatcher program to execute it. Multiple programs
75 can be attached at once with xdp_program__attach_multi(); they will be
76 sorted in order of their run priority, and execution from one program
77 to the next will proceed based on the chain call actions defined for
78 each program (see the Program metadata section below). Because the
79 loading process involves modifying the attach type of the program, the
80 attach functions only work with struct xdp_program objects that have
81 not yet been loaded into the kernel.
82 When using the attach functions to attach to an interface that already
85 has an XDP program loaded, libxdp will attempt to add the program to
86 the list of loaded programs. However, this may fail, either due to
87 missing kernel support, or because the already-attached program was not
88 loaded using a dispatcher compatible with libxdp. If the kernel support
89 for incremental attach (merged in kernel 5.10) is missing, the only way
90 to actually run multiple programs on a single interface is to attach
91 them all at the same time with xdp_program__attach_multi(). If the ex‐
92 isting program is not an XDP dispatcher, that program will have to be
93 detached from the interface before libxdp can attach a new one. This
94 can be done by calling xdp_program__detach() with a reference to the
95 loaded program; but note that this will of course break any application
96 relying on that other XDP program to be present.
97
100 To support multiple XDP programs on the same interface, libxdp uses two
101 pieces of metadata for each XDP program: Run priority and chain call
102 actions.
103 Run priority
106 This is the priority of the program and is a simple integer used to
107 sort programs when loading multiple programs onto the same interface.
108 Programs that wish to run early (such as a packet filter) should set
109 low values for this, while programs that want to run later (such as a
110 packet forwarder or counter) should set higher values. Note that later
111 programs are only run if the previous programs end with a return code
112 that is part of its chain call actions (see below). If not specified,
113 the default priority value is 50.
114 Chain call actions
117 These are the program return codes that the program indicate for pack‐
118 ets that should continue processing. If the program returns one of
119 these actions, later programs in the call chain will be run, whereas if
120 it returns any other action, processing will be interrupted, and the
121 XDP dispatcher will return the verdict immediately. If not set, this
122 defaults to just XDP_PASS, which is likely the value most programs
123 should use.
124 Specifying metadata
127 The metadata outlined above is specified as BTF information embedded in
128 the ELF file containing the XDP program. The xdp_helpers.h file shipped
129 with libxdp contains helper macros to include this information, which
130 can be used as follows:
131 #include <bpf/bpf_helpers.h>
133 #include <xdp/xdp_helpers.h>
134 struct {
136 __uint(priority, 10);
137 __uint(XDP_PASS, 1);
138 __uint(XDP_DROP, 1);
139 } XDP_RUN_CONFIG(my_xdp_func);
140 This example specifies that the XDP program in my_xdp_func should have
143 priority 10 and that its chain call actions are XDP_PASS and XDP_DROP.
144 In a source file with multiple XDP programs in the same file, a defini‐
145 tion like the above can be included for each program (main XDP func‐
146 tion). Any program that does not specify any config information will
147 use the default values outlined above.
148 Inspecting and modifying metadata
151 libxdp exposes the following functions that an application can use to
152 inspect and modify the metadata on an XDP program. Modification is only
153 possible before a program is attached on an interface. These functions
154 won't modify the BTF information itself, but the new values will be
155 stored as part of the program attachment.
156 unsigned int xdp_program__run_prio(const struct xdp_program *xdp_prog);
158 int xdp_program__set_run_prio(struct xdp_program *xdp_prog,
159 unsigned int run_prio);
160 bool xdp_program__chain_call_enabled(const struct xdp_program *xdp_prog,
161 enum xdp_action action);
162 int xdp_program__set_chain_call_enabled(struct xdp_program *prog,
163 unsigned int action,
164 bool enabled);
165 int xdp_program__print_chain_call_actions(const struct xdp_program *prog,
166 char *buf,
167 size_t buf_len);
168
171 To support multiple non-offloaded programs on the same network inter‐
172 face, libxdp uses a dispatcher program which is a small wrapper program
173 that will call each component program in turn, expect the return code,
174 and then chain call to the next program based on the chain call actions
175 of the previous program (see the Program metadata section above).
176 While applications using libxdp do not need to know the details of the
179 dispatcher program to just load an XDP program unto an interface,
180 libxdp does expose the dispatcher and its attached component programs,
181 which can be used to list the programs currently attached to an inter‐
182 face.
183 The structure used for this is struct xdp_multiprog, which can only be
186 constructed from the programs loaded on an interface based on ifindex.
187 The API for getting a multiprog reference and iterating through the at‐
188 tached programs looks like this:
189 struct xdp_multiprog *xdp_multiprog__get_from_ifindex(int ifindex);
191 struct xdp_program *xdp_multiprog__next_prog(const struct xdp_program *prog,
192 const struct xdp_multiprog *mp);
193 void xdp_multiprog__close(struct xdp_multiprog *mp);
194 int xdp_multiprog__detach(struct xdp_multiprog *mp, int ifindex);
195 enum xdp_attach_mode xdp_multiprog__attach_mode(const struct xdp_multiprog *mp);
196 struct xdp_program *xdp_multiprog__main_prog(const struct xdp_multiprog *mp);
197 struct xdp_program *xdp_multiprog__hw_prog(const struct xdp_multiprog *mp);
198 bool xdp_multiprog__is_legacy(const struct xdp_multiprog *mp);
199 If a non-offloaded program is attached to the interface which libxdp
202 doesn't recognise as a dispatcher program, an xdp_multiprog structure
203 will still be returned, and xdp_multiprog__is_legacy() will return true
204 for that program (note that this also holds true if only an offloaded
205 program is loaded). A reference to that (regular) XDP program can be
206 obtained by xdp_multiprog__main_prog(). If the program attached to the
207 interface is a dispatcher program, xdp_multiprog__main_prog() will re‐
208 turn a reference to the dispatcher program itself, which is mainly use‐
209 ful for obtaining other data about that program (such as the program
210 ID). A reference to an offloaded program can be acquired using xdp_mul‐
211 tiprog_hw_prog(). Function xdp_multiprog__attach_mode() returns the at‐
212 tach mode of the non-offloaded program, whether an offloaded program is
213 attached should be checked through xdp_multiprog_hw_prog().
214 Pinning in bpffs
217 The kernel will automatically detach component programs from the dis‐
218 patcher once the last reference to them disappears. To prevent this
219 from happening, libxdp will pin the component program references in
220 bpffs before attaching the dispatcher to the network interface. The
221 pathnames generated for pinning is as follows:
222 — /sys/fs/bpf/xdp/dispatch-IFINDEX-DID - dispatcher program for
225 IFINDEX with BPF program ID DID
226 — /sys/fs/bpf/xdp/dispatch-IFINDEX-DID/prog0-prog - component program
228 0, program reference
229 — /sys/fs/bpf/xdp/dispatch-IFINDEX-DID/prog0-link - component program
231 0, bpf_link reference
232 — /sys/fs/bpf/xdp/dispatch-IFINDEX-DID/prog1-prog - component program
234 1, program reference
235 — /sys/fs/bpf/xdp/dispatch-IFINDEX-DID/prog1-link - component program
237 1, bpf_link reference
238 — etc, up to ten component programs
240 If set, the LIBXDP_BPFFS environment variable will override the loca‐
243 tion of bpffs, but the xdp subdirectory is always used.
244
247 Libxdp implements helper functions for configuring AF_XDP sockets as
248 well as reading and writing packets from these sockets. AF_XDP sockets
249 can be used to redirect packets to user-space at high rates from an XDP
250 program. Note that this functionality used to reside in libbpf, but has
251 now been moved over to libxdp as it is a better fit for this library.
252 As of the 1.0 release of libbpf, the AF_XDP socket support will be re‐
253 moved and all future development will be performed in libxdp instead.
254 For an overview of AF_XDP sockets, please refer to this Linux Plumbers
257 paper (http://vger.kernel.org/lpc_net2018_talks/lpc18_pres_af_xdp_perf-
258 v3.pdf) and the documentation in the Linux kernel (Documentation/net‐
259 working/af_xdp.rst or https://www.kernel.org/doc/html/latest/network‐
260 ing/af_xdp.html).
261 For an example on how to use the interface, take a look at the sample
264 application in the Linux kernel source tree at samples/bpf/xdp‐
265 sock_user.c.
266 Control path
269 Libxdp provides helper functions for creating and destroying umems and
270 sockets as shown below. The first thing that a user generally wants to
271 do is to create a umem area. This is the area that will contain all
272 packets received and the ones that are going to be sent. After that,
273 AF_XDP sockets can be created tied to this umem. These can either be
274 sockets that have exclusive ownership of that umem through
275 xsk_socket__create() or shared with other sockets using
276 xsk_socket__create_shared. There is one option called
277 XSK_LIBBPF_FLAGS__INHIBIT_PROG_LOAD that can be set in the libxdp_flags
278 field (also called libbpf_flags for compatibility reasons). This will
279 make libxdp not load any XDP program or set and BPF maps which is a
280 must if users want to add their own XDP program.
281 int xsk_umem__create(struct xsk_umem **umem,
283 void *umem_area, __u64 size,
284 struct xsk_ring_prod *fill,
285 struct xsk_ring_cons *comp,
286 const struct xsk_umem_config *config);
287 int xsk_socket__create(struct xsk_socket **xsk,
288 const char *ifname, __u32 queue_id,
289 struct xsk_umem *umem,
290 struct xsk_ring_cons *rx,
291 struct xsk_ring_prod *tx,
292 const struct xsk_socket_config *config);
293 int xsk_socket__create_shared(struct xsk_socket **xsk_ptr,
294 const char *ifname,
295 __u32 queue_id, struct xsk_umem *umem,
296 struct xsk_ring_cons *rx,
297 struct xsk_ring_prod *tx,
298 struct xsk_ring_prod *fill,
299 struct xsk_ring_cons *comp,
300 const struct xsk_socket_config *config);
301 int xsk_umem__delete(struct xsk_umem *umem);
302 void xsk_socket__delete(struct xsk_socket *xsk);
303 There are also two helper function to get the file descriptor of a umem
306 or a socket. These are needed when using standard Linux syscalls such
307 as poll(), recvmsg(), sendto(), etc.
308 int xsk_umem__fd(const struct xsk_umem *umem);
310 int xsk_socket__fd(const struct xsk_socket *xsk);
311 The control path also provides two APIs for setting up AF_XDP sockets
314 when the process that is going to use the AF_XDP socket is non-privi‐
315 leged. These two functions perform the operations that require privi‐
316 leges and can be executed from some form of control process that has
317 the necessary privileges. The xsk_socket__create executed on the non-
318 privileged process will then skip these two steps. For an example on
319 how to use these, please take a look at https://github.com/tor‐
320 valds/linux/blob/master/samples/bpf/xdpsock_user.c at samples/bpf/xdp‐
321 sock_user.c and https://github.com/torvalds/linux/blob/master/sam‐
322 ples/bpf/xdpsock_ctrl_proc.c at samples/bpf/xdpsock_ctrl_proc.c in the
323 Linux kernel source tree.
324 int xsk_setup_xdp_prog(int ifindex, int *xsks_map_fd);
326 int xsk_socket__update_xskmap(struct xsk_socket *xsk, int xsks_map_fd);
327 Data path
330 For performance reasons, all the data path functions are static inline
331 functions found in the xsk.h header file so they can be optimized into
332 the target application binary for best possible performance. There are
333 four FIFO rings of two main types: producer rings (fill and Tx) and
334 consumer rings (Rx and completion). The producer rings use
335 xsk_ring_prod functions and consumer rings use xsk_ring_cons functions.
336 For producer rings, you start with reserving one or more slots in a
337 producer ring and then when they have been filled out, you submit them
338 so that the kernel will act on them. For a consumer ring, you peek if
339 there are any new packets in the ring and if so you can read them from
340 the ring. Once you are done reading them, you release them back to the
341 kernel so it can use them for new packets. There is also a cancel oper‐
342 ation for consumer rings if the application does not want to consume
343 all packets received with the peek operation.
344 __u32 xsk_ring_prod__reserve(struct xsk_ring_prod *prod, __u32 nb, __u32 *idx);
346 void xsk_ring_prod__submit(struct xsk_ring_prod *prod, __u32 nb);
347 __u32 xsk_ring_cons__peek(struct xsk_ring_cons *cons, __u32 nb, __u32 *idx);
348 void xsk_ring_cons__cancel(struct xsk_ring_cons *cons, __u32 nb);
349 void xsk_ring_cons__release(struct xsk_ring_cons *cons, __u32 nb);
350 The functions below are used for reading and writing the descriptors of
353 the rings. xsk_ring_prod__fill_addr() and xsk_ring_prod__tx_desc()
354 writes entries in the fill and Tx rings respectively, while
355 xsk_ring_cons__comp_addr and xsk_ring_cons__rx_desc reads entries from
356 the completion and Rx rings respectively. The idx is the parameter re‐
357 turned in the xsk_ring_prod__reserve or xsk_ring_cons__peek calls. To
358 advance to the next entry, simply do idx++.
359 __u64 *xsk_ring_prod__fill_addr(struct xsk_ring_prod *fill, __u32 idx);
361 struct xdp_desc *xsk_ring_prod__tx_desc(struct xsk_ring_prod *tx, __u32 idx);
362 const __u64 *xsk_ring_cons__comp_addr(const struct xsk_ring_cons *comp, __u32 idx);
363 const struct xdp_desc *xsk_ring_cons__rx_desc(const struct xsk_ring_cons *rx, __u32 idx);
364 The xsk_umem functions are used to get a pointer to the packet data it‐
367 self, always located inside the umem. In the default aligned mode, you
368 can get the addr variable straight from the Rx descriptor. But in un‐
369 aligned mode, you need to use the three last function below as the off‐
370 set used is carried in the upper 16 bits of the addr. Therefore, you
371 cannot use the addr straight from the descriptor in the unaligned case.
372 void *xsk_umem__get_data(void *umem_area, __u64 addr);
374 __u64 xsk_umem__extract_addr(__u64 addr);
375 __u64 xsk_umem__extract_offset(__u64 addr);
376 __u64 xsk_umem__add_offset_to_addr(__u64 addr);
377 There is one more function in the data path and that checks if the
380 need_wakeup flag is set. Use of this flag is highly encouraged and
381 should be enabled by setting XDP_USE_NEED_WAKEUP bit in the
382 xdp_bind_flags field that is provided to the xsk_socket_cre‐
383 ate_[shared]() calls. If this function returns true, then you need to
384 call recvmsg(), sendto(), or poll() depending on the situation.
385 recvmsg() if you are receiving, or sendto() if you are sending. poll()
386 can be used for both cases and provide the ability to sleep too, as
387 with any other socket. But note that poll is a slower operation than
388 the other two.
389 int xsk_ring_prod__needs_wakeup(const struct xsk_ring_prod *r);
391 For an example on how to use all these APIs, take a look at the sample
394 applications in the Linux kernel source tree at https://github.com/tor‐
395 valds/linux/blob/master/samples/bpf/xdpsock_user.c at samples/bpf/xdp‐
396 sock_user.c and https://github.com/torvalds/linux/blob/master/sam‐
397 ples/bpf/xsk_fwd.c at samples/bpf/xsk_fwd.c.
398
401 The features exposed by libxdp relies on certain kernel versions and
402 BPF features to work. To get the full benefit of all features, libxdp
403 needs to be used with kernel 5.10 or newer, unless the commits men‐
404 tioned below have been backported. However, libxdp will probe the ker‐
405 nel and transparently fall back to legacy loading procedures, so it is
406 possible to use the library with older versions, although some features
407 will be unavailable, as detailed below.
408 The ability to attach multiple BPF programs to a single interface re‐
411 lies on the kernel "BPF program extension" feature which was introduced
412 by commit be8704ff07d2 ("bpf: Introduce dynamic program extensions") in
413 the upstream kernel and first appeared in kernel release 5.6. To incre‐
414 mentally attach multiple programs, a further refinement added by commit
415 4a1e7c0c63e0 ("bpf: Support attaching freplace programs to multiple at‐
416 tach points") is needed; this first appeared in the upstream kernel
417 version 5.10. The functionality relies on the "BPF trampolines" feature
418 which is unfortunately only available on the x86_64 architecture. In
419 other words, kernels before 5.6 can only attach a single XDP program to
420 each interface, kernels 5.6+ can attach multiple programs if they are
421 all attached at the same time, and kernels 5.10 have full support for
422 XDP multiprog on x86_64. On other architectures, only a single program
423 can be attached to each interface.
424 To load AF_XDP programs, kernel support for AF_XDP sockets needs to be
427 included and enabled in the kernel build. In addition, when using
428 AF_XDP sockets, an XDP program is also loaded on the interface. The XDP
429 program used for this by libxdp requires the ability to do map lookups
430 into XSK maps, which was introduced with commit fada7fdc83c0 ("bpf: Al‐
431 low bpf_map_lookup_elem() on an xskmap") in kernel 5.3. This means that
432 the minimum required kernel version for using AF_XDP is kernel 5.3;
433 however, for the AF_XDP XDP program to co-exist with other programs,
434 the same constraints for multiprog applies as outlined above.
435 Note that some Linux distributions backport features to earlier kernel
438 versions, especially in enterprise kernels; for instance, Red Hat En‐
439 terprise Linux kernels include everything needed for libxdp to function
440 since RHEL 8.5.
441 Finally, XDP programs loaded using the multiprog facility must include
444 type information (using the BPF Type Format, BTF). To get this, compile
445 the programs with a recent version of Clang/LLVM (version 10+), and en‐
446 able debug information when compiling (using the -g option).
447
450 Please report any bugs on Github: https://github.com/xdp-project/xdp-
451 tools/issues
452
455 libxdp and this man page were written by Toke Høiland-Jørgensen. AF_XDP
456 support and documentation was contributed by Magnus Karlsson.
457v1.2.3 February 17, 2022 libxdp(3)