1 Linux Socket Filtering aka Berkeley Packet Filter (BPF)
2 =======================================================
7 Linux Socket Filtering (LSF) is derived from the Berkeley Packet Filter.
8 Though there are some distinct differences between the BSD and Linux
9 Kernel filtering, but when we speak of BPF or LSF in Linux context, we
10 mean the very same mechanism of filtering in the Linux kernel.
12 BPF allows a user-space program to attach a filter onto any socket and
13 allow or disallow certain types of data to come through the socket. LSF
14 follows exactly the same filter code structure as BSD's BPF, so referring
15 to the BSD bpf.4 manpage is very helpful in creating filters.
17 On Linux, BPF is much simpler than on BSD. One does not have to worry
18 about devices or anything like that. You simply create your filter code,
19 send it to the kernel via the SO_ATTACH_FILTER option and if your filter
20 code passes the kernel check on it, you then immediately begin filtering
23 You can also detach filters from your socket via the SO_DETACH_FILTER
24 option. This will probably not be used much since when you close a socket
25 that has a filter on it the filter is automagically removed. The other
26 less common case may be adding a different filter on the same socket where
27 you had another filter that is still running: the kernel takes care of
28 removing the old one and placing your new one in its place, assuming your
29 filter has passed the checks, otherwise if it fails the old filter will
30 remain on that socket.
32 SO_LOCK_FILTER option allows to lock the filter attached to a socket. Once
33 set, a filter cannot be removed or changed. This allows one process to
34 setup a socket, attach a filter, lock it then drop privileges and be
35 assured that the filter will be kept until the socket is closed.
37 The biggest user of this construct might be libpcap. Issuing a high-level
38 filter command like `tcpdump -i em1 port 22` passes through the libpcap
39 internal compiler that generates a structure that can eventually be loaded
40 via SO_ATTACH_FILTER to the kernel. `tcpdump -i em1 port 22 -ddd`
41 displays what is being placed into this structure.
43 Although we were only speaking about sockets here, BPF in Linux is used
44 in many more places. There's xt_bpf for netfilter, cls_bpf in the kernel
45 qdisc layer, SECCOMP-BPF (SECure COMPuting [1]), and lots of other places
46 such as team driver, PTP code, etc where BPF is being used.
48 [1] Documentation/userspace-api/seccomp_filter.rst
52 Steven McCanne and Van Jacobson. 1993. The BSD packet filter: a new
53 architecture for user-level packet capture. In Proceedings of the
54 USENIX Winter 1993 Conference Proceedings on USENIX Winter 1993
55 Conference Proceedings (USENIX'93). USENIX Association, Berkeley,
56 CA, USA, 2-2. [http://www.tcpdump.org/papers/bpf-usenix93.pdf]
61 User space applications include <linux/filter.h> which contains the
62 following relevant structures:
64 struct sock_filter { /* Filter block */
65 __u16 code; /* Actual filter code */
66 __u8 jt; /* Jump true */
67 __u8 jf; /* Jump false */
68 __u32 k; /* Generic multiuse field */
71 Such a structure is assembled as an array of 4-tuples, that contains
72 a code, jt, jf and k value. jt and jf are jump offsets and k a generic
73 value to be used for a provided code.
75 struct sock_fprog { /* Required for SO_ATTACH_FILTER. */
76 unsigned short len; /* Number of filter blocks */
77 struct sock_filter __user *filter;
80 For socket filtering, a pointer to this structure (as shown in
81 follow-up example) is being passed to the kernel through setsockopt(2).
86 #include <sys/socket.h>
87 #include <sys/types.h>
88 #include <arpa/inet.h>
89 #include <linux/if_ether.h>
92 /* From the example above: tcpdump -i em1 port 22 -dd */
93 struct sock_filter code[] = {
94 { 0x28, 0, 0, 0x0000000c },
95 { 0x15, 0, 8, 0x000086dd },
96 { 0x30, 0, 0, 0x00000014 },
97 { 0x15, 2, 0, 0x00000084 },
98 { 0x15, 1, 0, 0x00000006 },
99 { 0x15, 0, 17, 0x00000011 },
100 { 0x28, 0, 0, 0x00000036 },
101 { 0x15, 14, 0, 0x00000016 },
102 { 0x28, 0, 0, 0x00000038 },
103 { 0x15, 12, 13, 0x00000016 },
104 { 0x15, 0, 12, 0x00000800 },
105 { 0x30, 0, 0, 0x00000017 },
106 { 0x15, 2, 0, 0x00000084 },
107 { 0x15, 1, 0, 0x00000006 },
108 { 0x15, 0, 8, 0x00000011 },
109 { 0x28, 0, 0, 0x00000014 },
110 { 0x45, 6, 0, 0x00001fff },
111 { 0xb1, 0, 0, 0x0000000e },
112 { 0x48, 0, 0, 0x0000000e },
113 { 0x15, 2, 0, 0x00000016 },
114 { 0x48, 0, 0, 0x00000010 },
115 { 0x15, 0, 1, 0x00000016 },
116 { 0x06, 0, 0, 0x0000ffff },
117 { 0x06, 0, 0, 0x00000000 },
120 struct sock_fprog bpf = {
121 .len = ARRAY_SIZE(code),
125 sock = socket(PF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
127 /* ... bail out ... */
129 ret = setsockopt(sock, SOL_SOCKET, SO_ATTACH_FILTER, &bpf, sizeof(bpf));
131 /* ... bail out ... */
136 The above example code attaches a socket filter for a PF_PACKET socket
137 in order to let all IPv4/IPv6 packets with port 22 pass. The rest will
138 be dropped for this socket.
140 The setsockopt(2) call to SO_DETACH_FILTER doesn't need any arguments
141 and SO_LOCK_FILTER for preventing the filter to be detached, takes an
142 integer value with 0 or 1.
144 Note that socket filters are not restricted to PF_PACKET sockets only,
145 but can also be used on other socket families.
147 Summary of system calls:
149 * setsockopt(sockfd, SOL_SOCKET, SO_ATTACH_FILTER, &val, sizeof(val));
150 * setsockopt(sockfd, SOL_SOCKET, SO_DETACH_FILTER, &val, sizeof(val));
151 * setsockopt(sockfd, SOL_SOCKET, SO_LOCK_FILTER, &val, sizeof(val));
153 Normally, most use cases for socket filtering on packet sockets will be
154 covered by libpcap in high-level syntax, so as an application developer
155 you should stick to that. libpcap wraps its own layer around all that.
157 Unless i) using/linking to libpcap is not an option, ii) the required BPF
158 filters use Linux extensions that are not supported by libpcap's compiler,
159 iii) a filter might be more complex and not cleanly implementable with
160 libpcap's compiler, or iv) particular filter codes should be optimized
161 differently than libpcap's internal compiler does; then in such cases
162 writing such a filter "by hand" can be of an alternative. For example,
163 xt_bpf and cls_bpf users might have requirements that could result in
164 more complex filter code, or one that cannot be expressed with libpcap
165 (e.g. different return codes for various code paths). Moreover, BPF JIT
166 implementors may wish to manually write test cases and thus need low-level
167 access to BPF code as well.
169 BPF engine and instruction set
170 ------------------------------
172 Under tools/bpf/ there's a small helper tool called bpf_asm which can
173 be used to write low-level filters for example scenarios mentioned in the
174 previous section. Asm-like syntax mentioned here has been implemented in
175 bpf_asm and will be used for further explanations (instead of dealing with
176 less readable opcodes directly, principles are the same). The syntax is
177 closely modelled after Steven McCanne's and Van Jacobson's BPF paper.
179 The BPF architecture consists of the following basic elements:
183 A 32 bit wide accumulator
184 X 32 bit wide X register
185 M[] 16 x 32 bit wide misc registers aka "scratch memory
186 store", addressable from 0 to 15
188 A program, that is translated by bpf_asm into "opcodes" is an array that
189 consists of the following elements (as already mentioned):
191 op:16, jt:8, jf:8, k:32
193 The element op is a 16 bit wide opcode that has a particular instruction
194 encoded. jt and jf are two 8 bit wide jump targets, one for condition
195 "jump if true", the other one "jump if false". Eventually, element k
196 contains a miscellaneous argument that can be interpreted in different
197 ways depending on the given instruction in op.
199 The instruction set consists of load, store, branch, alu, miscellaneous
200 and return instructions that are also represented in bpf_asm syntax. This
201 table lists all bpf_asm instructions available resp. what their underlying
202 opcodes as defined in linux/filter.h stand for:
204 Instruction Addressing mode Description
206 ld 1, 2, 3, 4, 12 Load word into A
207 ldi 4 Load word into A
208 ldh 1, 2 Load half-word into A
209 ldb 1, 2 Load byte into A
210 ldx 3, 4, 5, 12 Load word into X
211 ldxi 4 Load word into X
212 ldxb 5 Load byte into X
214 st 3 Store A into M[]
215 stx 3 Store X into M[]
219 jeq 7, 8, 9, 10 Jump on A == <x>
220 jneq 9, 10 Jump on A != <x>
221 jne 9, 10 Jump on A != <x>
222 jlt 9, 10 Jump on A < <x>
223 jle 9, 10 Jump on A <= <x>
224 jgt 7, 8, 9, 10 Jump on A > <x>
225 jge 7, 8, 9, 10 Jump on A >= <x>
226 jset 7, 8, 9, 10 Jump on A & <x>
245 The next table shows addressing formats from the 2nd column:
247 Addressing mode Syntax Description
250 1 [k] BHW at byte offset k in the packet
251 2 [x + k] BHW at the offset X + k in the packet
252 3 M[k] Word at offset k in M[]
253 4 #k Literal value stored in k
254 5 4*([k]&0xf) Lower nibble * 4 at byte offset k in the packet
256 7 #k,Lt,Lf Jump to Lt if true, otherwise jump to Lf
257 8 x/%x,Lt,Lf Jump to Lt if true, otherwise jump to Lf
258 9 #k,Lt Jump to Lt if predicate is true
259 10 x/%x,Lt Jump to Lt if predicate is true
260 11 a/%a Accumulator A
261 12 extension BPF extension
263 The Linux kernel also has a couple of BPF extensions that are used along
264 with the class of load instructions by "overloading" the k argument with
265 a negative offset + a particular extension offset. The result of such BPF
266 extensions are loaded into A.
268 Possible BPF extensions are shown in the following table:
270 Extension Description
275 poff Payload start offset
276 ifidx skb->dev->ifindex
277 nla Netlink attribute of type X with offset A
278 nlan Nested Netlink attribute of type X with offset A
280 queue skb->queue_mapping
281 hatype skb->dev->type
283 cpu raw_smp_processor_id()
284 vlan_tci skb_vlan_tag_get(skb)
285 vlan_avail skb_vlan_tag_present(skb)
286 vlan_tpid skb->vlan_proto
289 These extensions can also be prefixed with '#'.
290 Examples for low-level BPF:
308 ** (Accelerated) VLAN w/ id 10:
315 ** icmp random packet sampling, 1 in 4
320 # get a random uint32 number
327 ** SECCOMP filter example:
329 ld [4] /* offsetof(struct seccomp_data, arch) */
330 jne #0xc000003e, bad /* AUDIT_ARCH_X86_64 */
331 ld [0] /* offsetof(struct seccomp_data, nr) */
332 jeq #15, good /* __NR_rt_sigreturn */
333 jeq #231, good /* __NR_exit_group */
334 jeq #60, good /* __NR_exit */
335 jeq #0, good /* __NR_read */
336 jeq #1, good /* __NR_write */
337 jeq #5, good /* __NR_fstat */
338 jeq #9, good /* __NR_mmap */
339 jeq #14, good /* __NR_rt_sigprocmask */
340 jeq #13, good /* __NR_rt_sigaction */
341 jeq #35, good /* __NR_nanosleep */
342 bad: ret #0 /* SECCOMP_RET_KILL_THREAD */
343 good: ret #0x7fff0000 /* SECCOMP_RET_ALLOW */
345 The above example code can be placed into a file (here called "foo"), and
346 then be passed to the bpf_asm tool for generating opcodes, output that xt_bpf
347 and cls_bpf understands and can directly be loaded with. Example with above
351 4,40 0 0 12,21 0 1 2054,6 0 0 4294967295,6 0 0 0,
353 In copy and paste C-like output:
356 { 0x28, 0, 0, 0x0000000c },
357 { 0x15, 0, 1, 0x00000806 },
358 { 0x06, 0, 0, 0xffffffff },
359 { 0x06, 0, 0, 0000000000 },
361 In particular, as usage with xt_bpf or cls_bpf can result in more complex BPF
362 filters that might not be obvious at first, it's good to test filters before
363 attaching to a live system. For that purpose, there's a small tool called
364 bpf_dbg under tools/bpf/ in the kernel source directory. This debugger allows
365 for testing BPF filters against given pcap files, single stepping through the
366 BPF code on the pcap's packets and to do BPF machine register dumps.
368 Starting bpf_dbg is trivial and just requires issuing:
372 In case input and output do not equal stdin/stdout, bpf_dbg takes an
373 alternative stdin source as a first argument, and an alternative stdout
374 sink as a second one, e.g. `./bpf_dbg test_in.txt test_out.txt`.
376 Other than that, a particular libreadline configuration can be set via
377 file "~/.bpf_dbg_init" and the command history is stored in the file
378 "~/.bpf_dbg_history".
380 Interaction in bpf_dbg happens through a shell that also has auto-completion
381 support (follow-up example commands starting with '>' denote bpf_dbg shell).
382 The usual workflow would be to ...
384 > load bpf 6,40 0 0 12,21 0 3 2048,48 0 0 23,21 0 1 1,6 0 0 65535,6 0 0 0
385 Loads a BPF filter from standard output of bpf_asm, or transformed via
386 e.g. `tcpdump -iem1 -ddd port 22 | tr '\n' ','`. Note that for JIT
387 debugging (next section), this command creates a temporary socket and
388 loads the BPF code into the kernel. Thus, this will also be useful for
392 Loads standard tcpdump pcap file.
396 Runs through all packets from a pcap to account how many passes and fails
397 the filter will generate. A limit of packets to traverse can be given.
401 l1: jeq #0x800, l2, l5
406 Prints out BPF code disassembly.
409 /* { op, jt, jf, k }, */
410 { 0x28, 0, 0, 0x0000000c },
411 { 0x15, 0, 3, 0x00000800 },
412 { 0x30, 0, 0, 0x00000017 },
413 { 0x15, 0, 1, 0x00000001 },
414 { 0x06, 0, 0, 0x0000ffff },
415 { 0x06, 0, 0, 0000000000 },
416 Prints out C-style BPF code dump.
419 breakpoint at: l0: ldh [12]
421 breakpoint at: l1: jeq #0x800, l2, l5
423 Sets breakpoints at particular BPF instructions. Issuing a `run` command
424 will walk through the pcap file continuing from the current packet and
425 break when a breakpoint is being hit (another `run` will continue from
426 the currently active breakpoint executing next instructions):
430 pc: [0] <-- program counter
431 code: [40] jt[0] jf[0] k[12] <-- plain BPF code of current instruction
432 curr: l0: ldh [12] <-- disassembly of current instruction
433 A: [00000000][0] <-- content of A (hex, decimal)
434 X: [00000000][0] <-- content of X (hex, decimal)
435 M[0,15]: [00000000][0] <-- folded content of M (hex, decimal)
436 -- packet dump -- <-- Current packet from pcap (hex)
438 0: 00 19 cb 55 55 a4 00 14 a4 43 78 69 08 06 00 01
439 16: 08 00 06 04 00 01 00 14 a4 43 78 69 0a 3b 01 26
440 32: 00 00 00 00 00 00 0a 3b 01 01
446 Prints currently set breakpoints.
449 Performs single stepping through the BPF program from the current pc
450 offset. Thus, on each step invocation, above register dump is issued.
451 This can go forwards and backwards in time, a plain `step` will break
452 on the next BPF instruction, thus +1. (No `run` needs to be issued here.)
455 Selects a given packet from the pcap file to continue from. Thus, on
456 the next `run` or `step`, the BPF program is being evaluated against
457 the user pre-selected packet. Numbering starts just as in Wireshark
467 The Linux kernel has a built-in BPF JIT compiler for x86_64, SPARC, PowerPC,
468 ARM, ARM64, MIPS and s390 and can be enabled through CONFIG_BPF_JIT. The JIT
469 compiler is transparently invoked for each attached filter from user space
470 or for internal kernel users if it has been previously enabled by root:
472 echo 1 > /proc/sys/net/core/bpf_jit_enable
474 For JIT developers, doing audits etc, each compile run can output the generated
475 opcode image into the kernel log via:
477 echo 2 > /proc/sys/net/core/bpf_jit_enable
479 Example output from dmesg:
481 [ 3389.935842] flen=6 proglen=70 pass=3 image=ffffffffa0069c8f
482 [ 3389.935847] JIT code: 00000000: 55 48 89 e5 48 83 ec 60 48 89 5d f8 44 8b 4f 68
483 [ 3389.935849] JIT code: 00000010: 44 2b 4f 6c 4c 8b 87 d8 00 00 00 be 0c 00 00 00
484 [ 3389.935850] JIT code: 00000020: e8 1d 94 ff e0 3d 00 08 00 00 75 16 be 17 00 00
485 [ 3389.935851] JIT code: 00000030: 00 e8 28 94 ff e0 83 f8 01 75 07 b8 ff ff 00 00
486 [ 3389.935852] JIT code: 00000040: eb 02 31 c0 c9 c3
488 When CONFIG_BPF_JIT_ALWAYS_ON is enabled, bpf_jit_enable is permanently set to 1 and
489 setting any other value than that will return in failure. This is even the case for
490 setting bpf_jit_enable to 2, since dumping the final JIT image into the kernel log
491 is discouraged and introspection through bpftool (under tools/bpf/bpftool/) is the
492 generally recommended approach instead.
494 In the kernel source tree under tools/bpf/, there's bpf_jit_disasm for
495 generating disassembly out of the kernel log's hexdump:
498 70 bytes emitted from JIT compiler (pass:3, flen:6)
499 ffffffffa0069c8f + <x>:
503 8: mov %rbx,-0x8(%rbp)
504 c: mov 0x68(%rdi),%r9d
505 10: sub 0x6c(%rdi),%r9d
506 14: mov 0xd8(%rdi),%r8
508 20: callq 0xffffffffe0ff9442
510 2a: jne 0x0000000000000042
512 31: callq 0xffffffffe0ff945e
514 39: jne 0x0000000000000042
516 40: jmp 0x0000000000000044
521 Issuing option `-o` will "annotate" opcodes to resulting assembler
522 instructions, which can be very useful for JIT developers:
524 # ./bpf_jit_disasm -o
525 70 bytes emitted from JIT compiler (pass:3, flen:6)
526 ffffffffa0069c8f + <x>:
533 8: mov %rbx,-0x8(%rbp)
535 c: mov 0x68(%rdi),%r9d
537 10: sub 0x6c(%rdi),%r9d
539 14: mov 0xd8(%rdi),%r8
543 20: callq 0xffffffffe0ff9442
547 2a: jne 0x0000000000000042
551 31: callq 0xffffffffe0ff945e
555 39: jne 0x0000000000000042
559 40: jmp 0x0000000000000044
568 For BPF JIT developers, bpf_jit_disasm, bpf_asm and bpf_dbg provides a useful
569 toolchain for developing and testing the kernel's JIT compiler.
573 Internally, for the kernel interpreter, a different instruction set
574 format with similar underlying principles from BPF described in previous
575 paragraphs is being used. However, the instruction set format is modelled
576 closer to the underlying architecture to mimic native instruction sets, so
577 that a better performance can be achieved (more details later). This new
578 ISA is called 'eBPF' or 'internal BPF' interchangeably. (Note: eBPF which
579 originates from [e]xtended BPF is not the same as BPF extensions! While
580 eBPF is an ISA, BPF extensions date back to classic BPF's 'overloading'
581 of BPF_LD | BPF_{B,H,W} | BPF_ABS instruction.)
583 It is designed to be JITed with one to one mapping, which can also open up
584 the possibility for GCC/LLVM compilers to generate optimized eBPF code through
585 an eBPF backend that performs almost as fast as natively compiled code.
587 The new instruction set was originally designed with the possible goal in
588 mind to write programs in "restricted C" and compile into eBPF with a optional
589 GCC/LLVM backend, so that it can just-in-time map to modern 64-bit CPUs with
590 minimal performance overhead over two steps, that is, C -> eBPF -> native code.
592 Currently, the new format is being used for running user BPF programs, which
593 includes seccomp BPF, classic socket filters, cls_bpf traffic classifier,
594 team driver's classifier for its load-balancing mode, netfilter's xt_bpf
595 extension, PTP dissector/classifier, and much more. They are all internally
596 converted by the kernel into the new instruction set representation and run
597 in the eBPF interpreter. For in-kernel handlers, this all works transparently
598 by using bpf_prog_create() for setting up the filter, resp.
599 bpf_prog_destroy() for destroying it. The macro
600 BPF_PROG_RUN(filter, ctx) transparently invokes eBPF interpreter or JITed
601 code to run the filter. 'filter' is a pointer to struct bpf_prog that we
602 got from bpf_prog_create(), and 'ctx' the given context (e.g.
603 skb pointer). All constraints and restrictions from bpf_check_classic() apply
604 before a conversion to the new layout is being done behind the scenes!
606 Currently, the classic BPF format is being used for JITing on most 32-bit
607 architectures, whereas x86-64, aarch64, s390x, powerpc64, sparc64, arm32 perform
608 JIT compilation from eBPF instruction set.
610 Some core changes of the new internal format:
612 - Number of registers increase from 2 to 10:
614 The old format had two registers A and X, and a hidden frame pointer. The
615 new layout extends this to be 10 internal registers and a read-only frame
616 pointer. Since 64-bit CPUs are passing arguments to functions via registers
617 the number of args from eBPF program to in-kernel function is restricted
618 to 5 and one register is used to accept return value from an in-kernel
619 function. Natively, x86_64 passes first 6 arguments in registers, aarch64/
620 sparcv9/mips64 have 7 - 8 registers for arguments; x86_64 has 6 callee saved
621 registers, and aarch64/sparcv9/mips64 have 11 or more callee saved registers.
623 Therefore, eBPF calling convention is defined as:
625 * R0 - return value from in-kernel function, and exit value for eBPF program
626 * R1 - R5 - arguments from eBPF program to in-kernel function
627 * R6 - R9 - callee saved registers that in-kernel function will preserve
628 * R10 - read-only frame pointer to access stack
630 Thus, all eBPF registers map one to one to HW registers on x86_64, aarch64,
631 etc, and eBPF calling convention maps directly to ABIs used by the kernel on
632 64-bit architectures.
634 On 32-bit architectures JIT may map programs that use only 32-bit arithmetic
635 and may let more complex programs to be interpreted.
637 R0 - R5 are scratch registers and eBPF program needs spill/fill them if
638 necessary across calls. Note that there is only one eBPF program (== one
639 eBPF main routine) and it cannot call other eBPF functions, it can only
640 call predefined in-kernel functions, though.
642 - Register width increases from 32-bit to 64-bit:
644 Still, the semantics of the original 32-bit ALU operations are preserved
645 via 32-bit subregisters. All eBPF registers are 64-bit with 32-bit lower
646 subregisters that zero-extend into 64-bit if they are being written to.
647 That behavior maps directly to x86_64 and arm64 subregister definition, but
648 makes other JITs more difficult.
650 32-bit architectures run 64-bit internal BPF programs via interpreter.
651 Their JITs may convert BPF programs that only use 32-bit subregisters into
652 native instruction set and let the rest being interpreted.
654 Operation is 64-bit, because on 64-bit architectures, pointers are also
655 64-bit wide, and we want to pass 64-bit values in/out of kernel functions,
656 so 32-bit eBPF registers would otherwise require to define register-pair
657 ABI, thus, there won't be able to use a direct eBPF register to HW register
658 mapping and JIT would need to do combine/split/move operations for every
659 register in and out of the function, which is complex, bug prone and slow.
660 Another reason is the use of atomic 64-bit counters.
662 - Conditional jt/jf targets replaced with jt/fall-through:
664 While the original design has constructs such as "if (cond) jump_true;
665 else jump_false;", they are being replaced into alternative constructs like
666 "if (cond) jump_true; /* else fall-through */".
668 - Introduces bpf_call insn and register passing convention for zero overhead
669 calls from/to other kernel functions:
671 Before an in-kernel function call, the internal BPF program needs to
672 place function arguments into R1 to R5 registers to satisfy calling
673 convention, then the interpreter will take them from registers and pass
674 to in-kernel function. If R1 - R5 registers are mapped to CPU registers
675 that are used for argument passing on given architecture, the JIT compiler
676 doesn't need to emit extra moves. Function arguments will be in the correct
677 registers and BPF_CALL instruction will be JITed as single 'call' HW
678 instruction. This calling convention was picked to cover common call
679 situations without performance penalty.
681 After an in-kernel function call, R1 - R5 are reset to unreadable and R0 has
682 a return value of the function. Since R6 - R9 are callee saved, their state
683 is preserved across the call.
685 For example, consider three C functions:
687 u64 f1() { return (*_f2)(1); }
688 u64 f2(u64 a) { return f3(a + 1, a); }
689 u64 f3(u64 a, u64 b) { return a - b; }
691 GCC can compile f1, f3 into x86_64:
702 Function f2 in eBPF may look like:
710 If f2 is JITed and the pointer stored to '_f2'. The calls f1 -> f2 -> f3 and
711 returns will be seamless. Without JIT, __bpf_prog_run() interpreter needs to
712 be used to call into f2.
714 For practical reasons all eBPF programs have only one argument 'ctx' which is
715 already placed into R1 (e.g. on __bpf_prog_run() startup) and the programs
716 can call kernel functions with up to 5 arguments. Calls with 6 or more arguments
717 are currently not supported, but these restrictions can be lifted if necessary
720 On 64-bit architectures all register map to HW registers one to one. For
721 example, x86_64 JIT compiler can map them as ...
735 ... since x86_64 ABI mandates rdi, rsi, rdx, rcx, r8, r9 for argument passing
736 and rbx, r12 - r15 are callee saved.
738 Then the following internal BPF pseudo-program:
740 bpf_mov R6, R1 /* save ctx */
746 bpf_mov R7, R0 /* save foo() return value */
747 bpf_mov R1, R6 /* restore ctx for next call */
756 After JIT to x86_64 may look like:
761 mov %rbx,-0x228(%rbp)
762 mov %r13,-0x220(%rbp)
777 mov -0x228(%rbp),%rbx
778 mov -0x220(%rbp),%r13
782 Which is in this example equivalent in C to:
784 u64 bpf_filter(u64 ctx)
786 return foo(ctx, 2, 3, 4, 5) + bar(ctx, 6, 7, 8, 9);
789 In-kernel functions foo() and bar() with prototype: u64 (*)(u64 arg1, u64
790 arg2, u64 arg3, u64 arg4, u64 arg5); will receive arguments in proper
791 registers and place their return value into '%rax' which is R0 in eBPF.
792 Prologue and epilogue are emitted by JIT and are implicit in the
793 interpreter. R0-R5 are scratch registers, so eBPF program needs to preserve
794 them across the calls as defined by calling convention.
796 For example the following program is invalid:
803 After the call the registers R1-R5 contain junk values and cannot be read.
804 An in-kernel eBPF verifier is used to validate internal BPF programs.
806 Also in the new design, eBPF is limited to 4096 insns, which means that any
807 program will terminate quickly and will only call a fixed number of kernel
808 functions. Original BPF and the new format are two operand instructions,
809 which helps to do one-to-one mapping between eBPF insn and x86 insn during JIT.
811 The input context pointer for invoking the interpreter function is generic,
812 its content is defined by a specific use case. For seccomp register R1 points
813 to seccomp_data, for converted BPF filters R1 points to a skb.
815 A program, that is translated internally consists of the following elements:
817 op:16, jt:8, jf:8, k:32 ==> op:8, dst_reg:4, src_reg:4, off:16, imm:32
819 So far 87 internal BPF instructions were implemented. 8-bit 'op' opcode field
820 has room for new instructions. Some of them may use 16/24/32 byte encoding. New
821 instructions must be multiple of 8 bytes to preserve backward compatibility.
823 Internal BPF is a general purpose RISC instruction set. Not every register and
824 every instruction are used during translation from original BPF to new format.
825 For example, socket filters are not using 'exclusive add' instruction, but
826 tracing filters may do to maintain counters of events, for example. Register R9
827 is not used by socket filters either, but more complex filters may be running
828 out of registers and would have to resort to spill/fill to stack.
830 Internal BPF can used as generic assembler for last step performance
831 optimizations, socket filters and seccomp are using it as assembler. Tracing
832 filters may use it as assembler to generate code from kernel. In kernel usage
833 may not be bounded by security considerations, since generated internal BPF code
834 may be optimizing internal code path and not being exposed to the user space.
835 Safety of internal BPF can come from a verifier (TBD). In such use cases as
836 described, it may be used as safe instruction set.
838 Just like the original BPF, the new format runs within a controlled environment,
839 is deterministic and the kernel can easily prove that. The safety of the program
840 can be determined in two steps: first step does depth-first-search to disallow
841 loops and other CFG validation; second step starts from the first insn and
842 descends all possible paths. It simulates execution of every insn and observes
843 the state change of registers and stack.
848 eBPF is reusing most of the opcode encoding from classic to simplify conversion
849 of classic BPF to eBPF. For arithmetic and jump instructions the 8-bit 'code'
850 field is divided into three parts:
852 +----------------+--------+--------------------+
853 | 4 bits | 1 bit | 3 bits |
854 | operation code | source | instruction class |
855 +----------------+--------+--------------------+
858 Three LSB bits store instruction class which is one of:
860 Classic BPF classes: eBPF classes:
862 BPF_LD 0x00 BPF_LD 0x00
863 BPF_LDX 0x01 BPF_LDX 0x01
864 BPF_ST 0x02 BPF_ST 0x02
865 BPF_STX 0x03 BPF_STX 0x03
866 BPF_ALU 0x04 BPF_ALU 0x04
867 BPF_JMP 0x05 BPF_JMP 0x05
868 BPF_RET 0x06 BPF_JMP32 0x06
869 BPF_MISC 0x07 BPF_ALU64 0x07
871 When BPF_CLASS(code) == BPF_ALU or BPF_JMP, 4th bit encodes source operand ...
876 * in classic BPF, this means:
878 BPF_SRC(code) == BPF_X - use register X as source operand
879 BPF_SRC(code) == BPF_K - use 32-bit immediate as source operand
881 * in eBPF, this means:
883 BPF_SRC(code) == BPF_X - use 'src_reg' register as source operand
884 BPF_SRC(code) == BPF_K - use 32-bit immediate as source operand
886 ... and four MSB bits store operation code.
888 If BPF_CLASS(code) == BPF_ALU or BPF_ALU64 [ in eBPF ], BPF_OP(code) is one of:
901 BPF_MOV 0xb0 /* eBPF only: mov reg to reg */
902 BPF_ARSH 0xc0 /* eBPF only: sign extending shift right */
903 BPF_END 0xd0 /* eBPF only: endianness conversion */
905 If BPF_CLASS(code) == BPF_JMP or BPF_JMP32 [ in eBPF ], BPF_OP(code) is one of:
907 BPF_JA 0x00 /* BPF_JMP only */
912 BPF_JNE 0x50 /* eBPF only: jump != */
913 BPF_JSGT 0x60 /* eBPF only: signed '>' */
914 BPF_JSGE 0x70 /* eBPF only: signed '>=' */
915 BPF_CALL 0x80 /* eBPF BPF_JMP only: function call */
916 BPF_EXIT 0x90 /* eBPF BPF_JMP only: function return */
917 BPF_JLT 0xa0 /* eBPF only: unsigned '<' */
918 BPF_JLE 0xb0 /* eBPF only: unsigned '<=' */
919 BPF_JSLT 0xc0 /* eBPF only: signed '<' */
920 BPF_JSLE 0xd0 /* eBPF only: signed '<=' */
922 So BPF_ADD | BPF_X | BPF_ALU means 32-bit addition in both classic BPF
923 and eBPF. There are only two registers in classic BPF, so it means A += X.
924 In eBPF it means dst_reg = (u32) dst_reg + (u32) src_reg; similarly,
925 BPF_XOR | BPF_K | BPF_ALU means A ^= imm32 in classic BPF and analogous
926 src_reg = (u32) src_reg ^ (u32) imm32 in eBPF.
928 Classic BPF is using BPF_MISC class to represent A = X and X = A moves.
929 eBPF is using BPF_MOV | BPF_X | BPF_ALU code instead. Since there are no
930 BPF_MISC operations in eBPF, the class 7 is used as BPF_ALU64 to mean
931 exactly the same operations as BPF_ALU, but with 64-bit wide operands
932 instead. So BPF_ADD | BPF_X | BPF_ALU64 means 64-bit addition, i.e.:
933 dst_reg = dst_reg + src_reg
935 Classic BPF wastes the whole BPF_RET class to represent a single 'ret'
936 operation. Classic BPF_RET | BPF_K means copy imm32 into return register
937 and perform function exit. eBPF is modeled to match CPU, so BPF_JMP | BPF_EXIT
938 in eBPF means function exit only. The eBPF program needs to store return
939 value into register R0 before doing a BPF_EXIT. Class 6 in eBPF is used as
940 BPF_JMP32 to mean exactly the same operations as BPF_JMP, but with 32-bit wide
941 operands for the comparisons instead.
943 For load and store instructions the 8-bit 'code' field is divided as:
945 +--------+--------+-------------------+
946 | 3 bits | 2 bits | 3 bits |
947 | mode | size | instruction class |
948 +--------+--------+-------------------+
951 Size modifier is one of ...
953 BPF_W 0x00 /* word */
954 BPF_H 0x08 /* half word */
955 BPF_B 0x10 /* byte */
956 BPF_DW 0x18 /* eBPF only, double word */
958 ... which encodes size of load/store operation:
963 DW - 8 byte (eBPF only)
965 Mode modifier is one of:
967 BPF_IMM 0x00 /* used for 32-bit mov in classic BPF and 64-bit in eBPF */
971 BPF_LEN 0x80 /* classic BPF only, reserved in eBPF */
972 BPF_MSH 0xa0 /* classic BPF only, reserved in eBPF */
973 BPF_XADD 0xc0 /* eBPF only, exclusive add */
975 eBPF has two non-generic instructions: (BPF_ABS | <size> | BPF_LD) and
976 (BPF_IND | <size> | BPF_LD) which are used to access packet data.
978 They had to be carried over from classic to have strong performance of
979 socket filters running in eBPF interpreter. These instructions can only
980 be used when interpreter context is a pointer to 'struct sk_buff' and
981 have seven implicit operands. Register R6 is an implicit input that must
982 contain pointer to sk_buff. Register R0 is an implicit output which contains
983 the data fetched from the packet. Registers R1-R5 are scratch registers
984 and must not be used to store the data across BPF_ABS | BPF_LD or
985 BPF_IND | BPF_LD instructions.
987 These instructions have implicit program exit condition as well. When
988 eBPF program is trying to access the data beyond the packet boundary,
989 the interpreter will abort the execution of the program. JIT compilers
990 therefore must preserve this property. src_reg and imm32 fields are
991 explicit inputs to these instructions.
995 BPF_IND | BPF_W | BPF_LD means:
997 R0 = ntohl(*(u32 *) (((struct sk_buff *) R6)->data + src_reg + imm32))
998 and R1 - R5 were scratched.
1000 Unlike classic BPF instruction set, eBPF has generic load/store operations:
1002 BPF_MEM | <size> | BPF_STX: *(size *) (dst_reg + off) = src_reg
1003 BPF_MEM | <size> | BPF_ST: *(size *) (dst_reg + off) = imm32
1004 BPF_MEM | <size> | BPF_LDX: dst_reg = *(size *) (src_reg + off)
1005 BPF_XADD | BPF_W | BPF_STX: lock xadd *(u32 *)(dst_reg + off16) += src_reg
1006 BPF_XADD | BPF_DW | BPF_STX: lock xadd *(u64 *)(dst_reg + off16) += src_reg
1008 Where size is one of: BPF_B or BPF_H or BPF_W or BPF_DW. Note that 1 and
1009 2 byte atomic increments are not supported.
1011 eBPF has one 16-byte instruction: BPF_LD | BPF_DW | BPF_IMM which consists
1012 of two consecutive 'struct bpf_insn' 8-byte blocks and interpreted as single
1013 instruction that loads 64-bit immediate value into a dst_reg.
1014 Classic BPF has similar instruction: BPF_LD | BPF_W | BPF_IMM which loads
1015 32-bit immediate value into a register.
1019 The safety of the eBPF program is determined in two steps.
1021 First step does DAG check to disallow loops and other CFG validation.
1022 In particular it will detect programs that have unreachable instructions.
1023 (though classic BPF checker allows them)
1025 Second step starts from the first insn and descends all possible paths.
1026 It simulates execution of every insn and observes the state change of
1027 registers and stack.
1029 At the start of the program the register R1 contains a pointer to context
1030 and has type PTR_TO_CTX.
1031 If verifier sees an insn that does R2=R1, then R2 has now type
1032 PTR_TO_CTX as well and can be used on the right hand side of expression.
1033 If R1=PTR_TO_CTX and insn is R2=R1+R1, then R2=SCALAR_VALUE,
1034 since addition of two valid pointers makes invalid pointer.
1035 (In 'secure' mode verifier will reject any type of pointer arithmetic to make
1036 sure that kernel addresses don't leak to unprivileged users)
1038 If register was never written to, it's not readable:
1041 will be rejected, since R2 is unreadable at the start of the program.
1043 After kernel function call, R1-R5 are reset to unreadable and
1044 R0 has a return type of the function.
1046 Since R6-R9 are callee saved, their state is preserved across the call.
1051 is a correct program. If there was R1 instead of R6, it would have
1054 load/store instructions are allowed only with registers of valid types, which
1055 are PTR_TO_CTX, PTR_TO_MAP, PTR_TO_STACK. They are bounds and alignment checked.
1059 bpf_xadd *(u32 *)(R1 + 3) += R2
1061 will be rejected, since R1 doesn't have a valid pointer type at the time of
1062 execution of instruction bpf_xadd.
1064 At the start R1 type is PTR_TO_CTX (a pointer to generic 'struct bpf_context')
1065 A callback is used to customize verifier to restrict eBPF program access to only
1066 certain fields within ctx structure with specified size and alignment.
1068 For example, the following insn:
1069 bpf_ld R0 = *(u32 *)(R6 + 8)
1070 intends to load a word from address R6 + 8 and store it into R0
1071 If R6=PTR_TO_CTX, via is_valid_access() callback the verifier will know
1072 that offset 8 of size 4 bytes can be accessed for reading, otherwise
1073 the verifier will reject the program.
1074 If R6=PTR_TO_STACK, then access should be aligned and be within
1075 stack bounds, which are [-MAX_BPF_STACK, 0). In this example offset is 8,
1076 so it will fail verification, since it's out of bounds.
1078 The verifier will allow eBPF program to read data from stack only after
1080 Classic BPF verifier does similar check with M[0-15] memory slots.
1082 bpf_ld R0 = *(u32 *)(R10 - 4)
1085 Though R10 is correct read-only register and has type PTR_TO_STACK
1086 and R10 - 4 is within stack bounds, there were no stores into that location.
1088 Pointer register spill/fill is tracked as well, since four (R6-R9)
1089 callee saved registers may not be enough for some programs.
1091 Allowed function calls are customized with bpf_verifier_ops->get_func_proto()
1092 The eBPF verifier will check that registers match argument constraints.
1093 After the call register R0 will be set to return type of the function.
1095 Function calls is a main mechanism to extend functionality of eBPF programs.
1096 Socket filters may let programs to call one set of functions, whereas tracing
1097 filters may allow completely different set.
1099 If a function made accessible to eBPF program, it needs to be thought through
1100 from safety point of view. The verifier will guarantee that the function is
1101 called with valid arguments.
1103 seccomp vs socket filters have different security restrictions for classic BPF.
1104 Seccomp solves this by two stage verifier: classic BPF verifier is followed
1105 by seccomp verifier. In case of eBPF one configurable verifier is shared for
1108 See details of eBPF verifier in kernel/bpf/verifier.c
1110 Register value tracking
1111 -----------------------
1112 In order to determine the safety of an eBPF program, the verifier must track
1113 the range of possible values in each register and also in each stack slot.
1114 This is done with 'struct bpf_reg_state', defined in include/linux/
1115 bpf_verifier.h, which unifies tracking of scalar and pointer values. Each
1116 register state has a type, which is either NOT_INIT (the register has not been
1117 written to), SCALAR_VALUE (some value which is not usable as a pointer), or a
1118 pointer type. The types of pointers describe their base, as follows:
1119 PTR_TO_CTX Pointer to bpf_context.
1120 CONST_PTR_TO_MAP Pointer to struct bpf_map. "Const" because arithmetic
1121 on these pointers is forbidden.
1122 PTR_TO_MAP_VALUE Pointer to the value stored in a map element.
1123 PTR_TO_MAP_VALUE_OR_NULL
1124 Either a pointer to a map value, or NULL; map accesses
1125 (see section 'eBPF maps', below) return this type,
1126 which becomes a PTR_TO_MAP_VALUE when checked != NULL.
1127 Arithmetic on these pointers is forbidden.
1128 PTR_TO_STACK Frame pointer.
1129 PTR_TO_PACKET skb->data.
1130 PTR_TO_PACKET_END skb->data + headlen; arithmetic forbidden.
1131 PTR_TO_SOCKET Pointer to struct bpf_sock_ops, implicitly refcounted.
1132 PTR_TO_SOCKET_OR_NULL
1133 Either a pointer to a socket, or NULL; socket lookup
1134 returns this type, which becomes a PTR_TO_SOCKET when
1135 checked != NULL. PTR_TO_SOCKET is reference-counted,
1136 so programs must release the reference through the
1137 socket release function before the end of the program.
1138 Arithmetic on these pointers is forbidden.
1139 However, a pointer may be offset from this base (as a result of pointer
1140 arithmetic), and this is tracked in two parts: the 'fixed offset' and 'variable
1141 offset'. The former is used when an exactly-known value (e.g. an immediate
1142 operand) is added to a pointer, while the latter is used for values which are
1143 not exactly known. The variable offset is also used in SCALAR_VALUEs, to track
1144 the range of possible values in the register.
1145 The verifier's knowledge about the variable offset consists of:
1146 * minimum and maximum values as unsigned
1147 * minimum and maximum values as signed
1148 * knowledge of the values of individual bits, in the form of a 'tnum': a u64
1149 'mask' and a u64 'value'. 1s in the mask represent bits whose value is unknown;
1150 1s in the value represent bits known to be 1. Bits known to be 0 have 0 in both
1151 mask and value; no bit should ever be 1 in both. For example, if a byte is read
1152 into a register from memory, the register's top 56 bits are known zero, while
1153 the low 8 are unknown - which is represented as the tnum (0x0; 0xff). If we
1154 then OR this with 0x40, we get (0x40; 0xbf), then if we add 1 we get (0x0;
1155 0x1ff), because of potential carries.
1157 Besides arithmetic, the register state can also be updated by conditional
1158 branches. For instance, if a SCALAR_VALUE is compared > 8, in the 'true' branch
1159 it will have a umin_value (unsigned minimum value) of 9, whereas in the 'false'
1160 branch it will have a umax_value of 8. A signed compare (with BPF_JSGT or
1161 BPF_JSGE) would instead update the signed minimum/maximum values. Information
1162 from the signed and unsigned bounds can be combined; for instance if a value is
1163 first tested < 8 and then tested s> 4, the verifier will conclude that the value
1164 is also > 4 and s< 8, since the bounds prevent crossing the sign boundary.
1166 PTR_TO_PACKETs with a variable offset part have an 'id', which is common to all
1167 pointers sharing that same variable offset. This is important for packet range
1168 checks: after adding a variable to a packet pointer register A, if you then copy
1169 it to another register B and then add a constant 4 to A, both registers will
1170 share the same 'id' but the A will have a fixed offset of +4. Then if A is
1171 bounds-checked and found to be less than a PTR_TO_PACKET_END, the register B is
1172 now known to have a safe range of at least 4 bytes. See 'Direct packet access',
1173 below, for more on PTR_TO_PACKET ranges.
1175 The 'id' field is also used on PTR_TO_MAP_VALUE_OR_NULL, common to all copies of
1176 the pointer returned from a map lookup. This means that when one copy is
1177 checked and found to be non-NULL, all copies can become PTR_TO_MAP_VALUEs.
1178 As well as range-checking, the tracked information is also used for enforcing
1179 alignment of pointer accesses. For instance, on most systems the packet pointer
1180 is 2 bytes after a 4-byte alignment. If a program adds 14 bytes to that to jump
1181 over the Ethernet header, then reads IHL and addes (IHL * 4), the resulting
1182 pointer will have a variable offset known to be 4n+2 for some n, so adding the 2
1183 bytes (NET_IP_ALIGN) gives a 4-byte alignment and so word-sized accesses through
1184 that pointer are safe.
1185 The 'id' field is also used on PTR_TO_SOCKET and PTR_TO_SOCKET_OR_NULL, common
1186 to all copies of the pointer returned from a socket lookup. This has similar
1187 behaviour to the handling for PTR_TO_MAP_VALUE_OR_NULL->PTR_TO_MAP_VALUE, but
1188 it also handles reference tracking for the pointer. PTR_TO_SOCKET implicitly
1189 represents a reference to the corresponding 'struct sock'. To ensure that the
1190 reference is not leaked, it is imperative to NULL-check the reference and in
1191 the non-NULL case, and pass the valid reference to the socket release function.
1193 Direct packet access
1194 --------------------
1195 In cls_bpf and act_bpf programs the verifier allows direct access to the packet
1196 data via skb->data and skb->data_end pointers.
1198 1: r4 = *(u32 *)(r1 +80) /* load skb->data_end */
1199 2: r3 = *(u32 *)(r1 +76) /* load skb->data */
1202 5: if r5 > r4 goto pc+16
1203 R1=ctx R3=pkt(id=0,off=0,r=14) R4=pkt_end R5=pkt(id=0,off=14,r=14) R10=fp
1204 6: r0 = *(u16 *)(r3 +12) /* access 12 and 13 bytes of the packet */
1206 this 2byte load from the packet is safe to do, since the program author
1207 did check 'if (skb->data + 14 > skb->data_end) goto err' at insn #5 which
1208 means that in the fall-through case the register R3 (which points to skb->data)
1209 has at least 14 directly accessible bytes. The verifier marks it
1210 as R3=pkt(id=0,off=0,r=14).
1211 id=0 means that no additional variables were added to the register.
1212 off=0 means that no additional constants were added.
1213 r=14 is the range of safe access which means that bytes [R3, R3 + 14) are ok.
1214 Note that R5 is marked as R5=pkt(id=0,off=14,r=14). It also points
1215 to the packet data, but constant 14 was added to the register, so
1216 it now points to 'skb->data + 14' and accessible range is [R5, R5 + 14 - 14)
1217 which is zero bytes.
1219 More complex packet access may look like:
1220 R0=inv1 R1=ctx R3=pkt(id=0,off=0,r=14) R4=pkt_end R5=pkt(id=0,off=14,r=14) R10=fp
1221 6: r0 = *(u8 *)(r3 +7) /* load 7th byte from the packet */
1222 7: r4 = *(u8 *)(r3 +12)
1224 9: r3 = *(u32 *)(r1 +76) /* load skb->data */
1232 17: r1 = *(u32 *)(r1 +80) /* load skb->data_end */
1233 18: if r2 > r1 goto pc+2
1234 R0=inv(id=0,umax_value=255,var_off=(0x0; 0xff)) R1=pkt_end R2=pkt(id=2,off=8,r=8) R3=pkt(id=2,off=0,r=8) R4=inv(id=0,umax_value=3570,var_off=(0x0; 0xfffe)) R5=pkt(id=0,off=14,r=14) R10=fp
1235 19: r1 = *(u8 *)(r3 +4)
1236 The state of the register R3 is R3=pkt(id=2,off=0,r=8)
1237 id=2 means that two 'r3 += rX' instructions were seen, so r3 points to some
1238 offset within a packet and since the program author did
1239 'if (r3 + 8 > r1) goto err' at insn #18, the safe range is [R3, R3 + 8).
1240 The verifier only allows 'add'/'sub' operations on packet registers. Any other
1241 operation will set the register state to 'SCALAR_VALUE' and it won't be
1242 available for direct packet access.
1243 Operation 'r3 += rX' may overflow and become less than original skb->data,
1244 therefore the verifier has to prevent that. So when it sees 'r3 += rX'
1245 instruction and rX is more than 16-bit value, any subsequent bounds-check of r3
1246 against skb->data_end will not give us 'range' information, so attempts to read
1247 through the pointer will give "invalid access to packet" error.
1248 Ex. after insn 'r4 = *(u8 *)(r3 +12)' (insn #7 above) the state of r4 is
1249 R4=inv(id=0,umax_value=255,var_off=(0x0; 0xff)) which means that upper 56 bits
1250 of the register are guaranteed to be zero, and nothing is known about the lower
1251 8 bits. After insn 'r4 *= 14' the state becomes
1252 R4=inv(id=0,umax_value=3570,var_off=(0x0; 0xfffe)), since multiplying an 8-bit
1253 value by constant 14 will keep upper 52 bits as zero, also the least significant
1254 bit will be zero as 14 is even. Similarly 'r2 >>= 48' will make
1255 R2=inv(id=0,umax_value=65535,var_off=(0x0; 0xffff)), since the shift is not sign
1256 extending. This logic is implemented in adjust_reg_min_max_vals() function,
1257 which calls adjust_ptr_min_max_vals() for adding pointer to scalar (or vice
1258 versa) and adjust_scalar_min_max_vals() for operations on two scalars.
1260 The end result is that bpf program author can access packet directly
1261 using normal C code as:
1262 void *data = (void *)(long)skb->data;
1263 void *data_end = (void *)(long)skb->data_end;
1264 struct eth_hdr *eth = data;
1265 struct iphdr *iph = data + sizeof(*eth);
1266 struct udphdr *udp = data + sizeof(*eth) + sizeof(*iph);
1268 if (data + sizeof(*eth) + sizeof(*iph) + sizeof(*udp) > data_end)
1270 if (eth->h_proto != htons(ETH_P_IP))
1272 if (iph->protocol != IPPROTO_UDP || iph->ihl != 5)
1274 if (udp->dest == 53 || udp->source == 9)
1276 which makes such programs easier to write comparing to LD_ABS insn
1277 and significantly faster.
1281 'maps' is a generic storage of different types for sharing data between kernel
1284 The maps are accessed from user space via BPF syscall, which has commands:
1285 - create a map with given type and attributes
1286 map_fd = bpf(BPF_MAP_CREATE, union bpf_attr *attr, u32 size)
1287 using attr->map_type, attr->key_size, attr->value_size, attr->max_entries
1288 returns process-local file descriptor or negative error
1290 - lookup key in a given map
1291 err = bpf(BPF_MAP_LOOKUP_ELEM, union bpf_attr *attr, u32 size)
1292 using attr->map_fd, attr->key, attr->value
1293 returns zero and stores found elem into value or negative error
1295 - create or update key/value pair in a given map
1296 err = bpf(BPF_MAP_UPDATE_ELEM, union bpf_attr *attr, u32 size)
1297 using attr->map_fd, attr->key, attr->value
1298 returns zero or negative error
1300 - find and delete element by key in a given map
1301 err = bpf(BPF_MAP_DELETE_ELEM, union bpf_attr *attr, u32 size)
1302 using attr->map_fd, attr->key
1304 - to delete map: close(fd)
1305 Exiting process will delete maps automatically
1307 userspace programs use this syscall to create/access maps that eBPF programs
1308 are concurrently updating.
1310 maps can have different types: hash, array, bloom filter, radix-tree, etc.
1312 The map is defined by:
1314 . max number of elements
1316 . value size in bytes
1320 The verifier does not actually walk all possible paths through the program. For
1321 each new branch to analyse, the verifier looks at all the states it's previously
1322 been in when at this instruction. If any of them contain the current state as a
1323 subset, the branch is 'pruned' - that is, the fact that the previous state was
1324 accepted implies the current state would be as well. For instance, if in the
1325 previous state, r1 held a packet-pointer, and in the current state, r1 holds a
1326 packet-pointer with a range as long or longer and at least as strict an
1327 alignment, then r1 is safe. Similarly, if r2 was NOT_INIT before then it can't
1328 have been used by any path from that point, so any value in r2 (including
1329 another NOT_INIT) is safe. The implementation is in the function regsafe().
1330 Pruning considers not only the registers but also the stack (and any spilled
1331 registers it may hold). They must all be safe for the branch to be pruned.
1332 This is implemented in states_equal().
1334 Understanding eBPF verifier messages
1335 ------------------------------------
1337 The following are few examples of invalid eBPF programs and verifier error
1338 messages as seen in the log:
1340 Program with unreachable instructions:
1341 static struct bpf_insn prog[] = {
1348 Program that reads uninitialized register:
1349 BPF_MOV64_REG(BPF_REG_0, BPF_REG_2),
1355 Program that doesn't initialize R0 before exiting:
1356 BPF_MOV64_REG(BPF_REG_2, BPF_REG_1),
1363 Program that accesses stack out of bounds:
1364 BPF_ST_MEM(BPF_DW, BPF_REG_10, 8, 0),
1367 0: (7a) *(u64 *)(r10 +8) = 0
1368 invalid stack off=8 size=8
1370 Program that doesn't initialize stack before passing its address into function:
1371 BPF_MOV64_REG(BPF_REG_2, BPF_REG_10),
1372 BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -8),
1373 BPF_LD_MAP_FD(BPF_REG_1, 0),
1374 BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem),
1381 invalid indirect read from stack off -8+0 size 8
1383 Program that uses invalid map_fd=0 while calling to map_lookup_elem() function:
1384 BPF_ST_MEM(BPF_DW, BPF_REG_10, -8, 0),
1385 BPF_MOV64_REG(BPF_REG_2, BPF_REG_10),
1386 BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -8),
1387 BPF_LD_MAP_FD(BPF_REG_1, 0),
1388 BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem),
1391 0: (7a) *(u64 *)(r10 -8) = 0
1396 fd 0 is not pointing to valid bpf_map
1398 Program that doesn't check return value of map_lookup_elem() before accessing
1400 BPF_ST_MEM(BPF_DW, BPF_REG_10, -8, 0),
1401 BPF_MOV64_REG(BPF_REG_2, BPF_REG_10),
1402 BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -8),
1403 BPF_LD_MAP_FD(BPF_REG_1, 0),
1404 BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem),
1405 BPF_ST_MEM(BPF_DW, BPF_REG_0, 0, 0),
1408 0: (7a) *(u64 *)(r10 -8) = 0
1413 5: (7a) *(u64 *)(r0 +0) = 0
1414 R0 invalid mem access 'map_value_or_null'
1416 Program that correctly checks map_lookup_elem() returned value for NULL, but
1417 accesses the memory with incorrect alignment:
1418 BPF_ST_MEM(BPF_DW, BPF_REG_10, -8, 0),
1419 BPF_MOV64_REG(BPF_REG_2, BPF_REG_10),
1420 BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -8),
1421 BPF_LD_MAP_FD(BPF_REG_1, 0),
1422 BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem),
1423 BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 1),
1424 BPF_ST_MEM(BPF_DW, BPF_REG_0, 4, 0),
1427 0: (7a) *(u64 *)(r10 -8) = 0
1432 5: (15) if r0 == 0x0 goto pc+1
1434 6: (7a) *(u64 *)(r0 +4) = 0
1435 misaligned access off 4 size 8
1437 Program that correctly checks map_lookup_elem() returned value for NULL and
1438 accesses memory with correct alignment in one side of 'if' branch, but fails
1439 to do so in the other side of 'if' branch:
1440 BPF_ST_MEM(BPF_DW, BPF_REG_10, -8, 0),
1441 BPF_MOV64_REG(BPF_REG_2, BPF_REG_10),
1442 BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -8),
1443 BPF_LD_MAP_FD(BPF_REG_1, 0),
1444 BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem),
1445 BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2),
1446 BPF_ST_MEM(BPF_DW, BPF_REG_0, 0, 0),
1448 BPF_ST_MEM(BPF_DW, BPF_REG_0, 0, 1),
1451 0: (7a) *(u64 *)(r10 -8) = 0
1456 5: (15) if r0 == 0x0 goto pc+2
1458 6: (7a) *(u64 *)(r0 +0) = 0
1461 from 5 to 8: R0=imm0 R10=fp
1462 8: (7a) *(u64 *)(r0 +0) = 1
1463 R0 invalid mem access 'imm'
1465 Program that performs a socket lookup then sets the pointer to NULL without
1468 BPF_MOV64_IMM(BPF_REG_2, 0),
1469 BPF_STX_MEM(BPF_W, BPF_REG_10, BPF_REG_2, -8),
1470 BPF_MOV64_REG(BPF_REG_2, BPF_REG_10),
1471 BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -8),
1472 BPF_MOV64_IMM(BPF_REG_3, 4),
1473 BPF_MOV64_IMM(BPF_REG_4, 0),
1474 BPF_MOV64_IMM(BPF_REG_5, 0),
1475 BPF_EMIT_CALL(BPF_FUNC_sk_lookup_tcp),
1476 BPF_MOV64_IMM(BPF_REG_0, 0),
1480 1: (63) *(u32 *)(r10 -8) = r2
1486 7: (85) call bpf_sk_lookup_tcp#65
1489 Unreleased reference id=1, alloc_insn=7
1491 Program that performs a socket lookup but does not NULL-check the returned
1493 BPF_MOV64_IMM(BPF_REG_2, 0),
1494 BPF_STX_MEM(BPF_W, BPF_REG_10, BPF_REG_2, -8),
1495 BPF_MOV64_REG(BPF_REG_2, BPF_REG_10),
1496 BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -8),
1497 BPF_MOV64_IMM(BPF_REG_3, 4),
1498 BPF_MOV64_IMM(BPF_REG_4, 0),
1499 BPF_MOV64_IMM(BPF_REG_5, 0),
1500 BPF_EMIT_CALL(BPF_FUNC_sk_lookup_tcp),
1504 1: (63) *(u32 *)(r10 -8) = r2
1510 7: (85) call bpf_sk_lookup_tcp#65
1512 Unreleased reference id=1, alloc_insn=7
1517 Next to the BPF toolchain, the kernel also ships a test module that contains
1518 various test cases for classic and internal BPF that can be executed against
1519 the BPF interpreter and JIT compiler. It can be found in lib/test_bpf.c and
1520 enabled via Kconfig:
1524 After the module has been built and installed, the test suite can be executed
1525 via insmod or modprobe against 'test_bpf' module. Results of the test cases
1526 including timings in nsec can be found in the kernel log (dmesg).
1531 Also trinity, the Linux syscall fuzzer, has built-in support for BPF and
1532 SECCOMP-BPF kernel fuzzing.
1537 The document was written in the hope that it is found useful and in order
1538 to give potential BPF hackers or security auditors a better overview of
1539 the underlying architecture.
1541 Jay Schulist <jschlst@samba.org>
1542 Daniel Borkmann <daniel@iogearbox.net>
1543 Alexei Starovoitov <ast@kernel.org>