2 # SPDX-License-Identifier: GPL-2.0-only
4 # Copyright (C) 2018-2019 Netronome Systems, Inc.
5 # Copyright (C) 2021 Isovalent, Inc.
7 # In case user attempts to run with Python 2.
8 from __future__ import print_function
15 helpersDocStart = 'Start of BPF helper function descriptions:'
17 class NoHelperFound(BaseException):
20 class NoSyscallCommandFound(BaseException):
23 class ParsingError(BaseException):
24 def __init__(self, line='<line not provided>', reader=None):
26 BaseException.__init__(self,
27 'Error at file offset %d, parsing line: %s' %
28 (reader.tell(), line))
30 BaseException.__init__(self, 'Error parsing line: %s' % line)
33 class APIElement(object):
35 An object representing the description of an aspect of the eBPF API.
36 @proto: prototype of the API symbol
37 @desc: textual description of the symbol
38 @ret: (optional) description of any associated return value
40 def __init__(self, proto='', desc='', ret=''):
46 class Helper(APIElement):
48 An object representing the description of an eBPF helper function.
49 @proto: function prototype of the helper function
50 @desc: textual description of the helper function
51 @ret: description of the return value of the helper function
53 def __init__(self, *args, **kwargs):
54 super().__init__(*args, **kwargs)
57 def proto_break_down(self):
59 Break down helper function protocol into smaller chunks: return type,
60 name, distincts arguments.
62 arg_re = re.compile(r'((\w+ )*?(\w+|...))( (\**)(\w+))?$')
64 proto_re = re.compile(r'(.+) (\**)(\w+)\(((([^,]+)(, )?){1,5})\)$')
66 capture = proto_re.match(self.proto)
67 res['ret_type'] = capture.group(1)
68 res['ret_star'] = capture.group(2)
69 res['name'] = capture.group(3)
72 args = capture.group(4).split(', ')
74 capture = arg_re.match(a)
76 'type' : capture.group(1),
77 'star' : capture.group(5),
78 'name' : capture.group(6)
84 class HeaderParser(object):
86 An object used to parse a file in order to extract the documentation of a
87 list of eBPF helper functions. All the helpers that can be retrieved are
88 stored as Helper object, in the self.helpers() array.
89 @filename: name of file to parse, usually include/uapi/linux/bpf.h in the
92 def __init__(self, filename):
93 self.reader = open(filename, 'r')
97 self.desc_unique_helpers = set()
98 self.define_unique_helpers = []
99 self.helper_enum_vals = {}
100 self.helper_enum_pos = {}
101 self.desc_syscalls = []
102 self.enum_syscalls = []
104 def parse_element(self):
105 proto = self.parse_symbol()
106 desc = self.parse_desc(proto)
107 ret = self.parse_ret(proto)
108 return APIElement(proto=proto, desc=desc, ret=ret)
110 def parse_helper(self):
111 proto = self.parse_proto()
112 desc = self.parse_desc(proto)
113 ret = self.parse_ret(proto)
114 return Helper(proto=proto, desc=desc, ret=ret)
116 def parse_symbol(self):
117 p = re.compile(r' \* ?(BPF\w+)$')
118 capture = p.match(self.line)
120 raise NoSyscallCommandFound
121 end_re = re.compile(r' \* ?NOTES$')
122 end = end_re.match(self.line)
124 raise NoSyscallCommandFound
125 self.line = self.reader.readline()
126 return capture.group(1)
128 def parse_proto(self):
129 # Argument can be of shape:
133 # - Same as above, with "const" and/or "struct" in front of type
134 # - "..." (undefined number of arguments, for bpf_trace_printk())
135 # There is at least one term ("void"), and at most five arguments.
136 p = re.compile(r' \* ?((.+) \**\w+\((((const )?(struct )?(\w+|\.\.\.)( \**\w+)?)(, )?){1,5}\))$')
137 capture = p.match(self.line)
140 self.line = self.reader.readline()
141 return capture.group(1)
143 def parse_desc(self, proto):
144 p = re.compile(r' \* ?(?:\t| {5,8})Description$')
145 capture = p.match(self.line)
147 raise Exception("No description section found for " + proto)
148 # Description can be several lines, some of them possibly empty, and it
149 # stops when another subsection title is met.
153 self.line = self.reader.readline()
154 if self.line == ' *\n':
157 p = re.compile(r' \* ?(?:\t| {5,8})(?:\t| {8})(.*)')
158 capture = p.match(self.line)
161 desc += capture.group(1) + '\n'
166 raise Exception("No description found for " + proto)
169 def parse_ret(self, proto):
170 p = re.compile(r' \* ?(?:\t| {5,8})Return$')
171 capture = p.match(self.line)
173 raise Exception("No return section found for " + proto)
174 # Return value description can be several lines, some of them possibly
175 # empty, and it stops when another subsection title is met.
179 self.line = self.reader.readline()
180 if self.line == ' *\n':
183 p = re.compile(r' \* ?(?:\t| {5,8})(?:\t| {8})(.*)')
184 capture = p.match(self.line)
187 ret += capture.group(1) + '\n'
192 raise Exception("No return found for " + proto)
195 def seek_to(self, target, help_message, discard_lines = 1):
197 offset = self.reader.read().find(target)
199 raise Exception(help_message)
200 self.reader.seek(offset)
201 self.reader.readline()
202 for _ in range(discard_lines):
203 self.reader.readline()
204 self.line = self.reader.readline()
206 def parse_desc_syscall(self):
207 self.seek_to('* DOC: eBPF Syscall Commands',
208 'Could not find start of eBPF syscall descriptions list')
211 command = self.parse_element()
212 self.commands.append(command)
213 self.desc_syscalls.append(command.proto)
215 except NoSyscallCommandFound:
218 def parse_enum_syscall(self):
219 self.seek_to('enum bpf_cmd {',
220 'Could not find start of bpf_cmd enum', 0)
221 # Searches for either one or more BPF\w+ enums
222 bpf_p = re.compile(r'\s*(BPF\w+)+')
223 # Searches for an enum entry assigned to another entry,
224 # for e.g. BPF_PROG_RUN = BPF_PROG_TEST_RUN, which is
225 # not documented hence should be skipped in check to
226 # determine if the right number of syscalls are documented
227 assign_p = re.compile(r'\s*(BPF\w+)\s*=\s*(BPF\w+)')
230 capture = assign_p.match(self.line)
232 # Skip line if an enum entry is assigned to another entry
233 self.line = self.reader.readline()
235 capture = bpf_p.match(self.line)
237 bpf_cmd_str += self.line
240 self.line = self.reader.readline()
241 # Find the number of occurences of BPF\w+
242 self.enum_syscalls = re.findall(r'(BPF\w+)+', bpf_cmd_str)
244 def parse_desc_helpers(self):
245 self.seek_to(helpersDocStart,
246 'Could not find start of eBPF helper descriptions list')
249 helper = self.parse_helper()
250 self.helpers.append(helper)
251 proto = helper.proto_break_down()
252 self.desc_unique_helpers.add(proto['name'])
253 except NoHelperFound:
256 def parse_define_helpers(self):
257 # Parse FN(...) in #define ___BPF_FUNC_MAPPER to compare later with the
258 # number of unique function names present in description and use the
259 # correct enumeration value.
260 # Note: seek_to(..) discards the first line below the target search text,
261 # resulting in FN(unspec, 0, ##ctx) being skipped and not added to
262 # self.define_unique_helpers.
263 self.seek_to('#define ___BPF_FUNC_MAPPER(FN, ctx...)',
264 'Could not find start of eBPF helper definition list')
265 # Searches for one FN(\w+) define or a backslash for newline
266 p = re.compile(r'\s*FN\((\w+), (\d+), ##ctx\)|\\\\')
270 capture = p.match(self.line)
272 fn_defines_str += self.line
273 helper_name = capture.expand(r'bpf_\1')
274 self.helper_enum_vals[helper_name] = int(capture.group(2))
275 self.helper_enum_pos[helper_name] = i
279 self.line = self.reader.readline()
280 # Find the number of occurences of FN(\w+)
281 self.define_unique_helpers = re.findall(r'FN\(\w+, \d+, ##ctx\)', fn_defines_str)
283 def validate_helpers(self):
286 seen_enum_vals = set()
288 for helper in self.helpers:
289 proto = helper.proto_break_down()
292 enum_val = self.helper_enum_vals[name]
293 enum_pos = self.helper_enum_pos[name]
295 raise Exception("Helper %s is missing from enum bpf_func_id" % name)
297 if name in seen_helpers:
298 if last_helper != name:
299 raise Exception("Helper %s has multiple descriptions which are not grouped together" % name)
302 # Enforce current practice of having the descriptions ordered
305 raise Exception("Helper %s (ID %d) comment order (#%d) must be aligned with its position (#%d) in enum bpf_func_id" % (name, enum_val, i + 1, enum_pos + 1))
306 if enum_val in seen_enum_vals:
307 raise Exception("Helper %s has duplicated value %d" % (name, enum_val))
309 seen_helpers.add(name)
311 seen_enum_vals.add(enum_val)
313 helper.enum_val = enum_val
317 self.parse_desc_syscall()
318 self.parse_enum_syscall()
319 self.parse_desc_helpers()
320 self.parse_define_helpers()
321 self.validate_helpers()
324 ###############################################################################
326 class Printer(object):
328 A generic class for printers. Printers should be created with an array of
329 Helper objects, and implement a way to print them in the desired fashion.
330 @parser: A HeaderParser with objects to print to standard output
332 def __init__(self, parser):
336 def print_header(self):
339 def print_footer(self):
342 def print_one(self, helper):
347 for elem in self.elements:
351 def elem_number_check(self, desc_unique_elem, define_unique_elem, type, instance):
353 Checks the number of helpers/syscalls documented within the header file
354 description with those defined as part of enum/macro and raise an
355 Exception if they don't match.
357 nr_desc_unique_elem = len(desc_unique_elem)
358 nr_define_unique_elem = len(define_unique_elem)
359 if nr_desc_unique_elem != nr_define_unique_elem:
361 The number of unique %s in description (%d) doesn\'t match the number of unique %s defined in %s (%d)
362 ''' % (type, nr_desc_unique_elem, type, instance, nr_define_unique_elem)
363 if nr_desc_unique_elem < nr_define_unique_elem:
364 # Function description is parsed until no helper is found (which can be due to
365 # misformatting). Hence, only print the first missing/misformatted helper/enum.
367 The description for %s is not present or formatted correctly.
368 ''' % (define_unique_elem[nr_desc_unique_elem])
369 raise Exception(exception_msg)
371 class PrinterRST(Printer):
373 A generic class for printers that print ReStructured Text. Printers should
374 be created with a HeaderParser object, and implement a way to print API
375 elements in the desired fashion.
376 @parser: A HeaderParser with objects to print to standard output
378 def __init__(self, parser):
381 def print_license(self):
383 .. Copyright (C) All BPF authors and contributors from 2014 to present.
384 .. See git log include/uapi/linux/bpf.h in kernel tree for details.
386 .. SPDX-License-Identifier: Linux-man-pages-copyleft
388 .. Please do not edit this file. It was generated from the documentation
389 .. located in file include/uapi/linux/bpf.h of the Linux kernel sources
390 .. (helpers description), and from scripts/bpf_doc.py in the same
391 .. repository (header and footer).
395 def print_elem(self, elem):
397 print('\tDescription')
398 # Do not strip all newline characters: formatted code at the end of
399 # a section must be followed by a blank line.
400 for line in re.sub('\n$', '', elem.desc, count=1).split('\n'):
401 print('{}{}'.format('\t\t' if line else '', line))
405 for line in elem.ret.rstrip().split('\n'):
406 print('{}{}'.format('\t\t' if line else '', line))
410 def get_kernel_version(self):
412 version = subprocess.run(['git', 'describe'], cwd=linuxRoot,
413 capture_output=True, check=True)
414 version = version.stdout.decode().rstrip()
417 version = subprocess.run(['make', 'kernelversion'], cwd=linuxRoot,
418 capture_output=True, check=True)
419 version = version.stdout.decode().rstrip()
422 return 'Linux {version}'.format(version=version)
424 def get_last_doc_update(self, delimiter):
426 cmd = ['git', 'log', '-1', '--pretty=format:%cs', '--no-patch',
428 '/{}/,/\\*\\//:include/uapi/linux/bpf.h'.format(delimiter)]
429 date = subprocess.run(cmd, cwd=linuxRoot,
430 capture_output=True, check=True)
431 return date.stdout.decode().rstrip()
435 class PrinterHelpersRST(PrinterRST):
437 A printer for dumping collected information about helpers as a ReStructured
438 Text page compatible with the rst2man program, which can be used to
439 generate a manual page for the helpers.
440 @parser: A HeaderParser with Helper objects to print to standard output
442 def __init__(self, parser):
443 self.elements = parser.helpers
444 self.elem_number_check(parser.desc_unique_helpers, parser.define_unique_helpers, 'helper', '___BPF_FUNC_MAPPER')
446 def print_header(self):
451 -------------------------------------------------------------------------------
452 list of eBPF helper functions
453 -------------------------------------------------------------------------------
462 The extended Berkeley Packet Filter (eBPF) subsystem consists in programs
463 written in a pseudo-assembly language, then attached to one of the several
464 kernel hooks and run in reaction of specific events. This framework differs
465 from the older, "classic" BPF (or "cBPF") in several aspects, one of them being
466 the ability to call special functions (or "helpers") from within a program.
467 These functions are restricted to a white-list of helpers defined in the
470 These helpers are used by eBPF programs to interact with the system, or with
471 the context in which they work. For instance, they can be used to print
472 debugging messages, to get the time since the system was booted, to interact
473 with eBPF maps, or to manipulate network packets. Since there are several eBPF
474 program types, and that they do not run in the same context, each program type
475 can only call a subset of those helpers.
477 Due to eBPF conventions, a helper can not have more than five arguments.
479 Internally, eBPF programs call directly into the compiled helper functions
480 without requiring any foreign-function interface. As a result, calling helpers
481 introduces no overhead, thus offering excellent performance.
483 This document is an attempt to list and document the helpers available to eBPF
484 developers. They are sorted by chronological order (the oldest helpers in the
490 kernelVersion = self.get_kernel_version()
491 lastUpdate = self.get_last_doc_update(helpersDocStart)
493 PrinterRST.print_license(self)
494 print(header.format(version=kernelVersion,
495 date_field = ':Date: ' if lastUpdate else '',
498 def print_footer(self):
503 Example usage for most of the eBPF helpers listed in this manual page are
504 available within the Linux kernel sources, at the following locations:
507 * *tools/testing/selftests/bpf/*
512 eBPF programs can have an associated license, passed along with the bytecode
513 instructions to the kernel when the programs are loaded. The format for that
514 string is identical to the one in use for kernel modules (Dual licenses, such
515 as "Dual BSD/GPL", may be used). Some helper functions are only accessible to
516 programs that are compatible with the GNU Privacy License (GPL).
518 In order to use such helpers, the eBPF program must be loaded with the correct
519 license string passed (via **attr**) to the **bpf**\\ () system call, and this
520 generally translates into the C source code of the program containing a line
521 similar to the following:
525 char ____license[] __attribute__((section("license"), used)) = "GPL";
530 This manual page is an effort to document the existing eBPF helper functions.
531 But as of this writing, the BPF sub-system is under heavy development. New eBPF
532 program or map types are added, along with new helper functions. Some helpers
533 are occasionally made available for additional program types. So in spite of
534 the efforts of the community, this page might not be up-to-date. If you want to
535 check by yourself what helper functions exist in your kernel, or what types of
536 programs they can support, here are some files among the kernel tree that you
537 may be interested in:
539 * *include/uapi/linux/bpf.h* is the main BPF header. It contains the full list
540 of all helper functions, as well as many other BPF definitions including most
541 of the flags, structs or constants used by the helpers.
542 * *net/core/filter.c* contains the definition of most network-related helper
543 functions, and the list of program types from which they can be used.
544 * *kernel/trace/bpf_trace.c* is the equivalent for most tracing program-related
546 * *kernel/bpf/verifier.c* contains the functions used to check that valid types
547 of eBPF maps are used with a given helper function.
548 * *kernel/bpf/* directory contains other files in which additional helpers are
549 defined (for cgroups, sockmaps, etc.).
550 * The bpftool utility can be used to probe the availability of helper functions
551 on the system (as well as supported program and map types, and a number of
552 other parameters). To do so, run **bpftool feature probe** (see
553 **bpftool-feature**\\ (8) for details). Add the **unprivileged** keyword to
554 list features available to unprivileged users.
556 Compatibility between helper functions and program types can generally be found
557 in the files where helper functions are defined. Look for the **struct
558 bpf_func_proto** objects and for functions returning them: these functions
559 contain a list of helpers that a given program type can call. Note that the
560 **default:** label of the **switch ... case** used to filter helpers can call
561 other functions, themselves allowing access to additional helpers. The
562 requirement for GPL license is also in those **struct bpf_func_proto**.
564 Compatibility between helper functions and map types can be found in the
565 **check_map_func_compatibility**\\ () function in file *kernel/bpf/verifier.c*.
567 Helper functions that invalidate the checks on **data** and **data_end**
568 pointers for network processing are listed in function
569 **bpf_helper_changes_pkt_data**\\ () in file *net/core/filter.c*.
578 **perf_event_open**\\ (2),
584 def print_proto(self, helper):
586 Format function protocol with bold and italics markers. This makes RST
587 file less readable, but gives nice results in the manual page.
589 proto = helper.proto_break_down()
591 print('**%s %s%s(' % (proto['ret_type'],
592 proto['ret_star'].replace('*', '\\*'),
597 for a in proto['args']:
598 one_arg = '{}{}'.format(comma, a['type'])
601 one_arg += ' {}**\\ '.format(a['star'].replace('*', '\\*'))
604 one_arg += '*{}*\\ **'.format(a['name'])
606 print(one_arg, end='')
610 def print_one(self, helper):
611 self.print_proto(helper)
612 self.print_elem(helper)
615 class PrinterSyscallRST(PrinterRST):
617 A printer for dumping collected information about the syscall API as a
618 ReStructured Text page compatible with the rst2man program, which can be
619 used to generate a manual page for the syscall.
620 @parser: A HeaderParser with APIElement objects to print to standard
623 def __init__(self, parser):
624 self.elements = parser.commands
625 self.elem_number_check(parser.desc_syscalls, parser.enum_syscalls, 'syscall', 'bpf_cmd')
627 def print_header(self):
632 -------------------------------------------------------------------------------
633 Perform a command on an extended BPF object
634 -------------------------------------------------------------------------------
641 PrinterRST.print_license(self)
644 def print_one(self, command):
645 print('**%s**' % (command.proto))
646 self.print_elem(command)
649 class PrinterHelpers(Printer):
651 A printer for dumping collected information about helpers as C header to
652 be included from BPF program.
653 @parser: A HeaderParser with Helper objects to print to standard output
655 def __init__(self, parser):
656 self.elements = parser.helpers
657 self.elem_number_check(parser.desc_unique_helpers, parser.define_unique_helpers, 'helper', '___BPF_FUNC_MAPPER')
660 'struct bpf_fib_lookup',
661 'struct bpf_sk_lookup',
662 'struct bpf_perf_event_data',
663 'struct bpf_perf_event_value',
664 'struct bpf_pidns_info',
665 'struct bpf_redir_neigh',
667 'struct bpf_sock_addr',
668 'struct bpf_sock_ops',
669 'struct bpf_sock_tuple',
670 'struct bpf_spin_lock',
672 'struct bpf_tcp_sock',
673 'struct bpf_tunnel_key',
674 'struct bpf_xfrm_state',
675 'struct linux_binprm',
677 'struct sk_reuseport_md',
683 'struct tcp_timewait_sock',
684 'struct tcp_request_sock',
687 'struct task_struct',
718 'struct bpf_fib_lookup',
719 'struct bpf_perf_event_data',
720 'struct bpf_perf_event_value',
721 'struct bpf_pidns_info',
722 'struct bpf_redir_neigh',
723 'struct bpf_sk_lookup',
725 'struct bpf_sock_addr',
726 'struct bpf_sock_ops',
727 'struct bpf_sock_tuple',
728 'struct bpf_spin_lock',
730 'struct bpf_tcp_sock',
731 'struct bpf_tunnel_key',
732 'struct bpf_xfrm_state',
733 'struct linux_binprm',
735 'struct sk_reuseport_md',
741 'struct tcp_timewait_sock',
742 'struct tcp_request_sock',
745 'struct task_struct',
755 'const struct bpf_dynptr',
768 'size_t': 'unsigned long',
769 'struct bpf_map': 'void',
770 'struct sk_buff': 'struct __sk_buff',
771 'const struct sk_buff': 'const struct __sk_buff',
772 'struct sk_msg_buff': 'struct sk_msg_md',
773 'struct xdp_buff': 'struct xdp_md',
775 # Helpers overloaded for different context types.
776 overloaded_helpers = [
777 'bpf_get_socket_cookie',
781 def print_header(self):
783 /* This is auto-generated file. See bpf_doc.py for details. */
785 /* Forward declarations of BPF structs */'''
788 for fwd in self.type_fwds:
792 def print_footer(self):
796 def map_type(self, t):
797 if t in self.known_types:
799 if t in self.mapped_types:
800 return self.mapped_types[t]
801 print("Unrecognized type '%s', please add it to known types!" % t,
807 def print_one(self, helper):
808 proto = helper.proto_break_down()
810 if proto['name'] in self.seen_helpers:
812 self.seen_helpers.add(proto['name'])
815 print(" * %s" % proto['name'])
818 # Do not strip all newline characters: formatted code at the end of
819 # a section must be followed by a blank line.
820 for line in re.sub('\n$', '', helper.desc, count=1).split('\n'):
821 print(' *{}{}'.format(' \t' if line else '', line))
826 for line in helper.ret.rstrip().split('\n'):
827 print(' *{}{}'.format(' \t' if line else '', line))
830 print('static %s %s(*%s)(' % (self.map_type(proto['ret_type']),
831 proto['ret_star'], proto['name']), end='')
833 for i, a in enumerate(proto['args']):
836 if proto['name'] in self.overloaded_helpers and i == 0:
839 one_arg = '{}{}'.format(comma, self.map_type(t))
842 one_arg += ' {}'.format(a['star'])
845 one_arg += '{}'.format(n)
847 print(one_arg, end='')
849 print(') = (void *) %d;' % helper.enum_val)
852 ###############################################################################
854 # If script is launched from scripts/ from kernel tree and can access
855 # ../include/uapi/linux/bpf.h, use it as a default name for the file to parse,
856 # otherwise the --filename argument will be required from the command line.
857 script = os.path.abspath(sys.argv[0])
858 linuxRoot = os.path.dirname(os.path.dirname(script))
859 bpfh = os.path.join(linuxRoot, 'include/uapi/linux/bpf.h')
862 'helpers': PrinterHelpersRST,
863 'syscall': PrinterSyscallRST,
866 argParser = argparse.ArgumentParser(description="""
867 Parse eBPF header file and generate documentation for the eBPF API.
868 The RST-formatted output produced can be turned into a manual page with the
871 argParser.add_argument('--header', action='store_true',
872 help='generate C header file')
873 if (os.path.isfile(bpfh)):
874 argParser.add_argument('--filename', help='path to include/uapi/linux/bpf.h',
877 argParser.add_argument('--filename', help='path to include/uapi/linux/bpf.h')
878 argParser.add_argument('target', nargs='?', default='helpers',
879 choices=printers.keys(), help='eBPF API target')
880 args = argParser.parse_args()
883 headerParser = HeaderParser(args.filename)
886 # Print formatted output to standard output.
888 if args.target != 'helpers':
889 raise NotImplementedError('Only helpers header generation is supported')
890 printer = PrinterHelpers(headerParser)
892 printer = printers[args.target](headerParser)