1#!/usr/bin/env python3 2# SPDX-License-Identifier: GPL-2.0-only 3# 4# Copyright (C) 2018-2019 Netronome Systems, Inc. 5# Copyright (C) 2021 Isovalent, Inc. 6 7# In case user attempts to run with Python 2. 8from __future__ import print_function 9 10import argparse 11import re 12import sys, os 13 14class NoHelperFound(BaseException): 15 pass 16 17class NoSyscallCommandFound(BaseException): 18 pass 19 20class ParsingError(BaseException): 21 def __init__(self, line='<line not provided>', reader=None): 22 if reader: 23 BaseException.__init__(self, 24 'Error at file offset %d, parsing line: %s' % 25 (reader.tell(), line)) 26 else: 27 BaseException.__init__(self, 'Error parsing line: %s' % line) 28 29 30class APIElement(object): 31 """ 32 An object representing the description of an aspect of the eBPF API. 33 @proto: prototype of the API symbol 34 @desc: textual description of the symbol 35 @ret: (optional) description of any associated return value 36 """ 37 def __init__(self, proto='', desc='', ret=''): 38 self.proto = proto 39 self.desc = desc 40 self.ret = ret 41 42 43class Helper(APIElement): 44 """ 45 An object representing the description of an eBPF helper function. 46 @proto: function prototype of the helper function 47 @desc: textual description of the helper function 48 @ret: description of the return value of the helper function 49 """ 50 def proto_break_down(self): 51 """ 52 Break down helper function protocol into smaller chunks: return type, 53 name, distincts arguments. 54 """ 55 arg_re = re.compile('((\w+ )*?(\w+|...))( (\**)(\w+))?$') 56 res = {} 57 proto_re = re.compile('(.+) (\**)(\w+)\(((([^,]+)(, )?){1,5})\)$') 58 59 capture = proto_re.match(self.proto) 60 res['ret_type'] = capture.group(1) 61 res['ret_star'] = capture.group(2) 62 res['name'] = capture.group(3) 63 res['args'] = [] 64 65 args = capture.group(4).split(', ') 66 for a in args: 67 capture = arg_re.match(a) 68 res['args'].append({ 69 'type' : capture.group(1), 70 'star' : capture.group(5), 71 'name' : capture.group(6) 72 }) 73 74 return res 75 76 77class HeaderParser(object): 78 """ 79 An object used to parse a file in order to extract the documentation of a 80 list of eBPF helper functions. All the helpers that can be retrieved are 81 stored as Helper object, in the self.helpers() array. 82 @filename: name of file to parse, usually include/uapi/linux/bpf.h in the 83 kernel tree 84 """ 85 def __init__(self, filename): 86 self.reader = open(filename, 'r') 87 self.line = '' 88 self.helpers = [] 89 self.commands = [] 90 91 def parse_element(self): 92 proto = self.parse_symbol() 93 desc = self.parse_desc() 94 ret = self.parse_ret() 95 return APIElement(proto=proto, desc=desc, ret=ret) 96 97 def parse_helper(self): 98 proto = self.parse_proto() 99 desc = self.parse_desc() 100 ret = self.parse_ret() 101 return Helper(proto=proto, desc=desc, ret=ret) 102 103 def parse_symbol(self): 104 p = re.compile(' \* ?(.+)$') 105 capture = p.match(self.line) 106 if not capture: 107 raise NoSyscallCommandFound 108 end_re = re.compile(' \* ?NOTES$') 109 end = end_re.match(self.line) 110 if end: 111 raise NoSyscallCommandFound 112 self.line = self.reader.readline() 113 return capture.group(1) 114 115 def parse_proto(self): 116 # Argument can be of shape: 117 # - "void" 118 # - "type name" 119 # - "type *name" 120 # - Same as above, with "const" and/or "struct" in front of type 121 # - "..." (undefined number of arguments, for bpf_trace_printk()) 122 # There is at least one term ("void"), and at most five arguments. 123 p = re.compile(' \* ?((.+) \**\w+\((((const )?(struct )?(\w+|\.\.\.)( \**\w+)?)(, )?){1,5}\))$') 124 capture = p.match(self.line) 125 if not capture: 126 raise NoHelperFound 127 self.line = self.reader.readline() 128 return capture.group(1) 129 130 def parse_desc(self): 131 p = re.compile(' \* ?(?:\t| {5,8})Description$') 132 capture = p.match(self.line) 133 if not capture: 134 # Helper can have empty description and we might be parsing another 135 # attribute: return but do not consume. 136 return '' 137 # Description can be several lines, some of them possibly empty, and it 138 # stops when another subsection title is met. 139 desc = '' 140 while True: 141 self.line = self.reader.readline() 142 if self.line == ' *\n': 143 desc += '\n' 144 else: 145 p = re.compile(' \* ?(?:\t| {5,8})(?:\t| {8})(.*)') 146 capture = p.match(self.line) 147 if capture: 148 desc += capture.group(1) + '\n' 149 else: 150 break 151 return desc 152 153 def parse_ret(self): 154 p = re.compile(' \* ?(?:\t| {5,8})Return$') 155 capture = p.match(self.line) 156 if not capture: 157 # Helper can have empty retval and we might be parsing another 158 # attribute: return but do not consume. 159 return '' 160 # Return value description can be several lines, some of them possibly 161 # empty, and it stops when another subsection title is met. 162 ret = '' 163 while True: 164 self.line = self.reader.readline() 165 if self.line == ' *\n': 166 ret += '\n' 167 else: 168 p = re.compile(' \* ?(?:\t| {5,8})(?:\t| {8})(.*)') 169 capture = p.match(self.line) 170 if capture: 171 ret += capture.group(1) + '\n' 172 else: 173 break 174 return ret 175 176 def seek_to(self, target, help_message): 177 self.reader.seek(0) 178 offset = self.reader.read().find(target) 179 if offset == -1: 180 raise Exception(help_message) 181 self.reader.seek(offset) 182 self.reader.readline() 183 self.reader.readline() 184 self.line = self.reader.readline() 185 186 def parse_syscall(self): 187 self.seek_to('* DOC: eBPF Syscall Commands', 188 'Could not find start of eBPF syscall descriptions list') 189 while True: 190 try: 191 command = self.parse_element() 192 self.commands.append(command) 193 except NoSyscallCommandFound: 194 break 195 196 def parse_helpers(self): 197 self.seek_to('* Start of BPF helper function descriptions:', 198 'Could not find start of eBPF helper descriptions list') 199 while True: 200 try: 201 helper = self.parse_helper() 202 self.helpers.append(helper) 203 except NoHelperFound: 204 break 205 206 def run(self): 207 self.parse_syscall() 208 self.parse_helpers() 209 self.reader.close() 210 211############################################################################### 212 213class Printer(object): 214 """ 215 A generic class for printers. Printers should be created with an array of 216 Helper objects, and implement a way to print them in the desired fashion. 217 @parser: A HeaderParser with objects to print to standard output 218 """ 219 def __init__(self, parser): 220 self.parser = parser 221 self.elements = [] 222 223 def print_header(self): 224 pass 225 226 def print_footer(self): 227 pass 228 229 def print_one(self, helper): 230 pass 231 232 def print_all(self): 233 self.print_header() 234 for elem in self.elements: 235 self.print_one(elem) 236 self.print_footer() 237 238 239class PrinterRST(Printer): 240 """ 241 A generic class for printers that print ReStructured Text. Printers should 242 be created with a HeaderParser object, and implement a way to print API 243 elements in the desired fashion. 244 @parser: A HeaderParser with objects to print to standard output 245 """ 246 def __init__(self, parser): 247 self.parser = parser 248 249 def print_license(self): 250 license = '''\ 251.. Copyright (C) All BPF authors and contributors from 2014 to present. 252.. See git log include/uapi/linux/bpf.h in kernel tree for details. 253.. 254.. %%%LICENSE_START(VERBATIM) 255.. Permission is granted to make and distribute verbatim copies of this 256.. manual provided the copyright notice and this permission notice are 257.. preserved on all copies. 258.. 259.. Permission is granted to copy and distribute modified versions of this 260.. manual under the conditions for verbatim copying, provided that the 261.. entire resulting derived work is distributed under the terms of a 262.. permission notice identical to this one. 263.. 264.. Since the Linux kernel and libraries are constantly changing, this 265.. manual page may be incorrect or out-of-date. The author(s) assume no 266.. responsibility for errors or omissions, or for damages resulting from 267.. the use of the information contained herein. The author(s) may not 268.. have taken the same level of care in the production of this manual, 269.. which is licensed free of charge, as they might when working 270.. professionally. 271.. 272.. Formatted or processed versions of this manual, if unaccompanied by 273.. the source, must acknowledge the copyright and authors of this work. 274.. %%%LICENSE_END 275.. 276.. Please do not edit this file. It was generated from the documentation 277.. located in file include/uapi/linux/bpf.h of the Linux kernel sources 278.. (helpers description), and from scripts/bpf_doc.py in the same 279.. repository (header and footer). 280''' 281 print(license) 282 283 def print_elem(self, elem): 284 if (elem.desc): 285 print('\tDescription') 286 # Do not strip all newline characters: formatted code at the end of 287 # a section must be followed by a blank line. 288 for line in re.sub('\n$', '', elem.desc, count=1).split('\n'): 289 print('{}{}'.format('\t\t' if line else '', line)) 290 291 if (elem.ret): 292 print('\tReturn') 293 for line in elem.ret.rstrip().split('\n'): 294 print('{}{}'.format('\t\t' if line else '', line)) 295 296 print('') 297 298 299class PrinterHelpersRST(PrinterRST): 300 """ 301 A printer for dumping collected information about helpers as a ReStructured 302 Text page compatible with the rst2man program, which can be used to 303 generate a manual page for the helpers. 304 @parser: A HeaderParser with Helper objects to print to standard output 305 """ 306 def __init__(self, parser): 307 self.elements = parser.helpers 308 309 def print_header(self): 310 header = '''\ 311=========== 312BPF-HELPERS 313=========== 314------------------------------------------------------------------------------- 315list of eBPF helper functions 316------------------------------------------------------------------------------- 317 318:Manual section: 7 319 320DESCRIPTION 321=========== 322 323The extended Berkeley Packet Filter (eBPF) subsystem consists in programs 324written in a pseudo-assembly language, then attached to one of the several 325kernel hooks and run in reaction of specific events. This framework differs 326from the older, "classic" BPF (or "cBPF") in several aspects, one of them being 327the ability to call special functions (or "helpers") from within a program. 328These functions are restricted to a white-list of helpers defined in the 329kernel. 330 331These helpers are used by eBPF programs to interact with the system, or with 332the context in which they work. For instance, they can be used to print 333debugging messages, to get the time since the system was booted, to interact 334with eBPF maps, or to manipulate network packets. Since there are several eBPF 335program types, and that they do not run in the same context, each program type 336can only call a subset of those helpers. 337 338Due to eBPF conventions, a helper can not have more than five arguments. 339 340Internally, eBPF programs call directly into the compiled helper functions 341without requiring any foreign-function interface. As a result, calling helpers 342introduces no overhead, thus offering excellent performance. 343 344This document is an attempt to list and document the helpers available to eBPF 345developers. They are sorted by chronological order (the oldest helpers in the 346kernel at the top). 347 348HELPERS 349======= 350''' 351 PrinterRST.print_license(self) 352 print(header) 353 354 def print_footer(self): 355 footer = ''' 356EXAMPLES 357======== 358 359Example usage for most of the eBPF helpers listed in this manual page are 360available within the Linux kernel sources, at the following locations: 361 362* *samples/bpf/* 363* *tools/testing/selftests/bpf/* 364 365LICENSE 366======= 367 368eBPF programs can have an associated license, passed along with the bytecode 369instructions to the kernel when the programs are loaded. The format for that 370string is identical to the one in use for kernel modules (Dual licenses, such 371as "Dual BSD/GPL", may be used). Some helper functions are only accessible to 372programs that are compatible with the GNU General Public License (GNU GPL). 373 374In order to use such helpers, the eBPF program must be loaded with the correct 375license string passed (via **attr**) to the **bpf**\ () system call, and this 376generally translates into the C source code of the program containing a line 377similar to the following: 378 379:: 380 381 char ____license[] __attribute__((section("license"), used)) = "GPL"; 382 383IMPLEMENTATION 384============== 385 386This manual page is an effort to document the existing eBPF helper functions. 387But as of this writing, the BPF sub-system is under heavy development. New eBPF 388program or map types are added, along with new helper functions. Some helpers 389are occasionally made available for additional program types. So in spite of 390the efforts of the community, this page might not be up-to-date. If you want to 391check by yourself what helper functions exist in your kernel, or what types of 392programs they can support, here are some files among the kernel tree that you 393may be interested in: 394 395* *include/uapi/linux/bpf.h* is the main BPF header. It contains the full list 396 of all helper functions, as well as many other BPF definitions including most 397 of the flags, structs or constants used by the helpers. 398* *net/core/filter.c* contains the definition of most network-related helper 399 functions, and the list of program types from which they can be used. 400* *kernel/trace/bpf_trace.c* is the equivalent for most tracing program-related 401 helpers. 402* *kernel/bpf/verifier.c* contains the functions used to check that valid types 403 of eBPF maps are used with a given helper function. 404* *kernel/bpf/* directory contains other files in which additional helpers are 405 defined (for cgroups, sockmaps, etc.). 406* The bpftool utility can be used to probe the availability of helper functions 407 on the system (as well as supported program and map types, and a number of 408 other parameters). To do so, run **bpftool feature probe** (see 409 **bpftool-feature**\ (8) for details). Add the **unprivileged** keyword to 410 list features available to unprivileged users. 411 412Compatibility between helper functions and program types can generally be found 413in the files where helper functions are defined. Look for the **struct 414bpf_func_proto** objects and for functions returning them: these functions 415contain a list of helpers that a given program type can call. Note that the 416**default:** label of the **switch ... case** used to filter helpers can call 417other functions, themselves allowing access to additional helpers. The 418requirement for GPL license is also in those **struct bpf_func_proto**. 419 420Compatibility between helper functions and map types can be found in the 421**check_map_func_compatibility**\ () function in file *kernel/bpf/verifier.c*. 422 423Helper functions that invalidate the checks on **data** and **data_end** 424pointers for network processing are listed in function 425**bpf_helper_changes_pkt_data**\ () in file *net/core/filter.c*. 426 427SEE ALSO 428======== 429 430**bpf**\ (2), 431**bpftool**\ (8), 432**cgroups**\ (7), 433**ip**\ (8), 434**perf_event_open**\ (2), 435**sendmsg**\ (2), 436**socket**\ (7), 437**tc-bpf**\ (8)''' 438 print(footer) 439 440 def print_proto(self, helper): 441 """ 442 Format function protocol with bold and italics markers. This makes RST 443 file less readable, but gives nice results in the manual page. 444 """ 445 proto = helper.proto_break_down() 446 447 print('**%s %s%s(' % (proto['ret_type'], 448 proto['ret_star'].replace('*', '\\*'), 449 proto['name']), 450 end='') 451 452 comma = '' 453 for a in proto['args']: 454 one_arg = '{}{}'.format(comma, a['type']) 455 if a['name']: 456 if a['star']: 457 one_arg += ' {}**\ '.format(a['star'].replace('*', '\\*')) 458 else: 459 one_arg += '** ' 460 one_arg += '*{}*\\ **'.format(a['name']) 461 comma = ', ' 462 print(one_arg, end='') 463 464 print(')**') 465 466 def print_one(self, helper): 467 self.print_proto(helper) 468 self.print_elem(helper) 469 470 471class PrinterSyscallRST(PrinterRST): 472 """ 473 A printer for dumping collected information about the syscall API as a 474 ReStructured Text page compatible with the rst2man program, which can be 475 used to generate a manual page for the syscall. 476 @parser: A HeaderParser with APIElement objects to print to standard 477 output 478 """ 479 def __init__(self, parser): 480 self.elements = parser.commands 481 482 def print_header(self): 483 header = '''\ 484=== 485bpf 486=== 487------------------------------------------------------------------------------- 488Perform a command on an extended BPF object 489------------------------------------------------------------------------------- 490 491:Manual section: 2 492 493COMMANDS 494======== 495''' 496 PrinterRST.print_license(self) 497 print(header) 498 499 def print_one(self, command): 500 print('**%s**' % (command.proto)) 501 self.print_elem(command) 502 503 504class PrinterHelpers(Printer): 505 """ 506 A printer for dumping collected information about helpers as C header to 507 be included from BPF program. 508 @parser: A HeaderParser with Helper objects to print to standard output 509 """ 510 def __init__(self, parser): 511 self.elements = parser.helpers 512 513 type_fwds = [ 514 'struct bpf_fib_lookup', 515 'struct bpf_sk_lookup', 516 'struct bpf_perf_event_data', 517 'struct bpf_perf_event_value', 518 'struct bpf_pidns_info', 519 'struct bpf_redir_neigh', 520 'struct bpf_sock', 521 'struct bpf_sock_addr', 522 'struct bpf_sock_ops', 523 'struct bpf_sock_tuple', 524 'struct bpf_spin_lock', 525 'struct bpf_sysctl', 526 'struct bpf_tcp_sock', 527 'struct bpf_tunnel_key', 528 'struct bpf_xfrm_state', 529 'struct linux_binprm', 530 'struct pt_regs', 531 'struct sk_reuseport_md', 532 'struct sockaddr', 533 'struct tcphdr', 534 'struct seq_file', 535 'struct tcp6_sock', 536 'struct tcp_sock', 537 'struct tcp_timewait_sock', 538 'struct tcp_request_sock', 539 'struct udp6_sock', 540 'struct task_struct', 541 542 'struct __sk_buff', 543 'struct sk_msg_md', 544 'struct xdp_md', 545 'struct path', 546 'struct btf_ptr', 547 'struct inode', 548 'struct socket', 549 'struct file', 550 'struct bpf_timer', 551 ] 552 known_types = { 553 '...', 554 'void', 555 'const void', 556 'char', 557 'const char', 558 'int', 559 'long', 560 'unsigned long', 561 562 '__be16', 563 '__be32', 564 '__wsum', 565 566 'struct bpf_fib_lookup', 567 'struct bpf_perf_event_data', 568 'struct bpf_perf_event_value', 569 'struct bpf_pidns_info', 570 'struct bpf_redir_neigh', 571 'struct bpf_sk_lookup', 572 'struct bpf_sock', 573 'struct bpf_sock_addr', 574 'struct bpf_sock_ops', 575 'struct bpf_sock_tuple', 576 'struct bpf_spin_lock', 577 'struct bpf_sysctl', 578 'struct bpf_tcp_sock', 579 'struct bpf_tunnel_key', 580 'struct bpf_xfrm_state', 581 'struct linux_binprm', 582 'struct pt_regs', 583 'struct sk_reuseport_md', 584 'struct sockaddr', 585 'struct tcphdr', 586 'struct seq_file', 587 'struct tcp6_sock', 588 'struct tcp_sock', 589 'struct tcp_timewait_sock', 590 'struct tcp_request_sock', 591 'struct udp6_sock', 592 'struct task_struct', 593 'struct path', 594 'struct btf_ptr', 595 'struct inode', 596 'struct socket', 597 'struct file', 598 'struct bpf_timer', 599 } 600 mapped_types = { 601 'u8': '__u8', 602 'u16': '__u16', 603 'u32': '__u32', 604 'u64': '__u64', 605 's8': '__s8', 606 's16': '__s16', 607 's32': '__s32', 608 's64': '__s64', 609 'size_t': 'unsigned long', 610 'struct bpf_map': 'void', 611 'struct sk_buff': 'struct __sk_buff', 612 'const struct sk_buff': 'const struct __sk_buff', 613 'struct sk_msg_buff': 'struct sk_msg_md', 614 'struct xdp_buff': 'struct xdp_md', 615 } 616 # Helpers overloaded for different context types. 617 overloaded_helpers = [ 618 'bpf_get_socket_cookie', 619 'bpf_sk_assign', 620 ] 621 622 def print_header(self): 623 header = '''\ 624/* This is auto-generated file. See bpf_doc.py for details. */ 625 626/* Forward declarations of BPF structs */''' 627 628 print(header) 629 for fwd in self.type_fwds: 630 print('%s;' % fwd) 631 print('') 632 633 def print_footer(self): 634 footer = '' 635 print(footer) 636 637 def map_type(self, t): 638 if t in self.known_types: 639 return t 640 if t in self.mapped_types: 641 return self.mapped_types[t] 642 print("Unrecognized type '%s', please add it to known types!" % t, 643 file=sys.stderr) 644 sys.exit(1) 645 646 seen_helpers = set() 647 648 def print_one(self, helper): 649 proto = helper.proto_break_down() 650 651 if proto['name'] in self.seen_helpers: 652 return 653 self.seen_helpers.add(proto['name']) 654 655 print('/*') 656 print(" * %s" % proto['name']) 657 print(" *") 658 if (helper.desc): 659 # Do not strip all newline characters: formatted code at the end of 660 # a section must be followed by a blank line. 661 for line in re.sub('\n$', '', helper.desc, count=1).split('\n'): 662 print(' *{}{}'.format(' \t' if line else '', line)) 663 664 if (helper.ret): 665 print(' *') 666 print(' * Returns') 667 for line in helper.ret.rstrip().split('\n'): 668 print(' *{}{}'.format(' \t' if line else '', line)) 669 670 print(' */') 671 print('static %s %s(*%s)(' % (self.map_type(proto['ret_type']), 672 proto['ret_star'], proto['name']), end='') 673 comma = '' 674 for i, a in enumerate(proto['args']): 675 t = a['type'] 676 n = a['name'] 677 if proto['name'] in self.overloaded_helpers and i == 0: 678 t = 'void' 679 n = 'ctx' 680 one_arg = '{}{}'.format(comma, self.map_type(t)) 681 if n: 682 if a['star']: 683 one_arg += ' {}'.format(a['star']) 684 else: 685 one_arg += ' ' 686 one_arg += '{}'.format(n) 687 comma = ', ' 688 print(one_arg, end='') 689 690 print(') = (void *) %d;' % len(self.seen_helpers)) 691 print('') 692 693############################################################################### 694 695# If script is launched from scripts/ from kernel tree and can access 696# ../include/uapi/linux/bpf.h, use it as a default name for the file to parse, 697# otherwise the --filename argument will be required from the command line. 698script = os.path.abspath(sys.argv[0]) 699linuxRoot = os.path.dirname(os.path.dirname(script)) 700bpfh = os.path.join(linuxRoot, 'include/uapi/linux/bpf.h') 701 702printers = { 703 'helpers': PrinterHelpersRST, 704 'syscall': PrinterSyscallRST, 705} 706 707argParser = argparse.ArgumentParser(description=""" 708Parse eBPF header file and generate documentation for the eBPF API. 709The RST-formatted output produced can be turned into a manual page with the 710rst2man utility. 711""") 712argParser.add_argument('--header', action='store_true', 713 help='generate C header file') 714if (os.path.isfile(bpfh)): 715 argParser.add_argument('--filename', help='path to include/uapi/linux/bpf.h', 716 default=bpfh) 717else: 718 argParser.add_argument('--filename', help='path to include/uapi/linux/bpf.h') 719argParser.add_argument('target', nargs='?', default='helpers', 720 choices=printers.keys(), help='eBPF API target') 721args = argParser.parse_args() 722 723# Parse file. 724headerParser = HeaderParser(args.filename) 725headerParser.run() 726 727# Print formatted output to standard output. 728if args.header: 729 if args.target != 'helpers': 730 raise NotImplementedError('Only helpers header generation is supported') 731 printer = PrinterHelpers(headerParser) 732else: 733 printer = printers[args.target](headerParser) 734printer.print_all() 735