1.\" Copyright (c) 1995 FreeBSD Inc. 2.\" All rights reserved. 3.\" 4.\" Redistribution and use in source and binary forms, with or without 5.\" modification, are permitted provided that the following conditions 6.\" are met: 7.\" 1. Redistributions of source code must retain the above copyright 8.\" notice, this list of conditions and the following disclaimer. 9.\" 2. Redistributions in binary form must reproduce the above copyright 10.\" notice, this list of conditions and the following disclaimer in the 11.\" documentation and/or other materials provided with the distribution. 12.\" 13.\" THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 14.\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 15.\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 16.\" ARE DISCLAIMED. IN NO EVENT SHALL [your name] OR CONTRIBUTORS BE LIABLE 17.\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 18.\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 19.\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 20.\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 21.\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 22.\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 23.\" SUCH DAMAGE. 24.\" 25.\" 26.Dd December 14, 1995 27.Dt STYLE 9 28.Os 29.Sh NAME 30.Nm style 31.Nd "kernel source file style guide" 32.Sh DESCRIPTION 33This file specifies the preferred style for kernel source files in the 34.Fx 35source tree. It is also a guide for preferred userland code style. 36.Bd -literal 37/* 38 * Style guide for FreeBSD. Based on the CSRG's KNF (Kernel Normal Form). 39 * 40 * @(#)style 1.14 (Berkeley) 4/28/95 41 * $FreeBSD$ 42 */ 43 44/* 45 * VERY important single-line comments look like this. 46 */ 47 48/* Most single-line comments look like this. */ 49 50/* 51 * Multi-line comments look like this. Make them real sentences. Fill 52 * them so they look like real paragraphs. 53 */ 54.Ed 55.Pp 56After any copyright header, there is a blank line, and the 57.Va rcsid 58for source files. 59Version control system ID tags should only exist once in a file 60(unlike this one). 61Non-C/C++ source files follow the example above, while C/C++ source files 62follow the one below. 63All VCS (version control system) revision identification from files obtained 64from elsewhere should be maintained, including, where applicable, multiple IDs 65showing a file's history. 66In general, keep the IDs intact, including any `$'s. 67There is no reason to add "From" in front of foreign VCS IDs. 68Most 69.No non- Ns Fx 70VCS IDs should be indented by a tab if in a comment. 71.Bd -literal 72#include <sys/cdefs.h> 73__RCSID("@(#)style 1.14 (Berkeley) 4/28/95") 74__FBSDID("$FreeBSD$") 75.Ed 76.Pp 77Leave another blank line before the header files. 78.Pp 79Kernel include files (i.e. sys/*.h) come first; normally, include 80<sys/types.h> 81OR <sys/param.h>, but not both. <sys/types.h> includes <sys/cdefs.h>, 82and it's okay to depend on that. 83.Bd -literal 84#include <sys/types.h> /* Non-local includes in angle brackets. */ 85.Ed 86.Pp 87For a network program, put the network include files next. 88.Bd -literal 89#include <net/if.h> 90#include <net/if_dl.h> 91#include <net/route.h> 92#include <netinet/in.h> 93#include <protocols/rwhod.h> 94.Ed 95.Pp 96Leave a blank line before the next group, the /usr include files, 97which should be sorted alphabetically by name. 98.Bd -literal 99#include <stdio.h> 100.Ed 101.Pp 102Global pathnames are defined in /usr/include/paths.h. Pathnames local 103to the program go in pathnames.h in the local directory. 104.Bd -literal 105#include <paths.h> 106.Ed 107.Pp 108Leave another blank line before the user include files. 109.Bd -literal 110#include "pathnames.h" /* Local includes in double quotes. */ 111.Ed 112.Pp 113Do not #define or declare names in the implementation namespace except 114for implementing application interfaces. 115.Pp 116The names of 117.Dq Li unsafe 118macros (ones that have side effects), and the names of macros for 119manifest constants, are all in uppercase. 120The expansions of expression-like macros are either a single token 121or have outer parentheses. 122Put a single tab character between the 123.Ql #define 124and the macro name. 125If a macro is an inline expansion of a function, the function name is 126all in lowercase and the macro has the same name all in uppercase. 127.\" XXX the above conflicts with ANSI style where the names are the 128.\" same and you #undef the macro (if any) to get the function. 129.\" It is not followed for MALLOC(), and not very common if inline 130.\" functions are used. 131If a 132macro needs more than a single line, use braces 133.Sq ( \&{ 134and 135.Sq \&} ) . 136Right-justify the 137backslashes; it makes it easier to read. 138If the macro encapsulates a compound statement, enclose it in a 139.Dq Li do 140loop, 141so that it can safely be used in 142.Dq Li if 143statements. 144Any final statement-terminating semicolon should be 145supplied by the macro invocation rather than the macro, to make parsing easier 146for pretty-printers and editors. 147.Bd -literal 148#define MACRO(x, y) do { \e 149 variable = (x) + (y); \e 150 (y) += 2; \e 151} while(0) 152.Ed 153.Pp 154Enumeration values are all uppercase. 155.Bd -literal 156enum enumtype { ONE, TWO } et; 157.Ed 158.Pp 159When declaring variables in structures, declare them sorted by use, then 160by size, and then by alphabetical order. The first category normally 161doesn't apply, but there are exceptions. Each one gets its own line. 162Put a tab after the first word, i.e. use 163.Ql int^Ix; 164and 165.Ql struct^Ifoo *x; . 166.Pp 167Major structures should be declared at the top of the file in which they 168are used, or in separate header files if they are used in multiple 169source files. Use of the structures should be by separate declarations 170and should be "extern" if they are declared in a header file. 171.Bd -literal 172struct foo { 173 struct foo *next; /* List of active foo */ 174 struct mumble amumble; /* Comment for mumble */ 175 int bar; 176}; 177struct foo *foohead; /* Head of global foo list */ 178.Ed 179.Pp 180Use 181.Xr queue 3 182macros rather than rolling your own lists, whenever possible. Thus, 183the previous example would be better written: 184.Bd -literal 185#include <sys/queue.h> 186struct foo { 187 LIST_ENTRY(foo) link; /* Queue macro glue for foo lists */ 188 struct mumble amumble; /* Comment for mumble */ 189 int bar; 190}; 191LIST_HEAD(, foo) foohead; /* Head of global foo list */ 192.Ed 193.Pp 194Avoid using typedefs for structure types. This makes it impossible 195for applications to use pointers to such a structure opaquely, which 196is both possible and beneficial when using an ordinary struct tag. 197When convention requires a typedef, make its name match the struct 198tag. Avoid typedefs ending in 199.Dq Li \&_t , 200except as specified in Standard C or by 201.Tn POSIX . 202.Bd -literal 203/* Make the structure name match the typedef. */ 204typedef struct bar { 205 int level; 206} BAR; 207.Ed 208.Pp 209All functions are prototyped somewhere. 210.Pp 211Function prototypes for private functions (i.e. functions not used 212elsewhere) go at the top of the first source module. Functions 213local to one source module should be declared 214.Ql static . 215.Pp 216Functions used from other parts of the kernel are prototyped in the 217relevant include file. 218.Pp 219Functions that are used locally in more than one module go into a 220separate header file, e.g.\& 221.Pa extern.h . 222.Pp 223Only use the __P macro from the include file <sys/cdefs.h> if the source 224file in general is (to be) compilable with a K&R Old Testament compiler. 225Use of the __P macro in new code is discouraged, although modifications 226to existing files should be consistent with that file's conventions. 227.Pp 228In general code can be considered 229.Dq new code 230when it makes up about 50% or more of the file[s] involved. This is enough 231to break precedents in the existing code and use the current style guidelines. 232.Pp 233The kernel has a name associated with parameter types, e.g., in the kernel 234use: 235.Bd -literal 236void function(int fd); 237.Ed 238.Pp 239In header files visible to userland applications, prototypes that are 240visible must use either protected names or no names with the types. It 241is preferable to use protected names. 242E.g., use: 243.Bd -literal 244void function(int); 245.Ed 246.Pp 247or: 248.Bd -literal 249void function(int _fd); 250.Ed 251.Pp 252Prototypes may have an extra space after a tab to enable function names 253to line up: 254.Bd -literal 255static char *function(int _arg, const char *_arg2, struct foo *_arg3, 256 struct bar *_arg4); 257static void usage(void); 258 259/* 260 * All major routines should have a comment briefly describing what 261 * they do. The comment before the "main" routine should describe 262 * what the program does. 263 */ 264int 265main(int argc, char *argv[]) 266{ 267 long num; 268 int ch; 269 char *ep; 270 271.Ed 272.Pp 273For consistency, getopt should be used to parse options. Options 274should be sorted in the getopt call and the switch statement, unless 275parts of the switch cascade. Elements in a switch statement that 276cascade should have a FALLTHROUGH comment. Numerical arguments 277should be checked for accuracy. Code that cannot be reached should 278have a NOTREACHED comment. 279.Bd -literal 280 while ((ch = getopt(argc, argv, "abn:")) != -1) 281 switch (ch) { /* Indent the switch. */ 282 case 'a': /* Don't indent the case. */ 283 aflag = 1; 284 /* FALLTHROUGH */ 285 case 'b': 286 bflag = 1; 287 break; 288 case 'n': 289 num = strtol(optarg, &ep, 10); 290 if (num <= 0 || *ep != '\e0') { 291 warnx("illegal number, -n argument -- %s", 292 optarg); 293 usage(); 294 } 295 break; 296 case '?': 297 default: 298 usage(); 299 /* NOTREACHED */ 300 } 301 argc -= optind; 302 argv += optind; 303 304.Ed 305.Pp 306Space after keywords (if, while, for, return, switch). No braces are 307used for control statements with zero or only a single statement unless that 308statement is more than a single line in which case they are permitted. 309Forever loops are done with for's, not while's. 310.Bd -literal 311 for (p = buf; *p != '\e0'; ++p) 312 ; /* nothing */ 313 for (;;) 314 stmt; 315 for (;;) { 316 z = a + really + long + statement + that + needs + 317 two lines + gets + indented + four + spaces + 318 on + the + second + and + subsequent + lines; 319 } 320 for (;;) { 321 if (cond) 322 stmt; 323 } 324 if (val != NULL) 325 val = realloc(val, newsize); 326.Ed 327.Pp 328Parts of a for loop may be left empty. Don't put declarations 329inside blocks unless the routine is unusually complicated. 330.Bd -literal 331 for (; cnt < 15; cnt++) { 332 stmt1; 333 stmt2; 334 } 335.Ed 336.Pp 337Indentation is an 8 character tab. 338Second level indents are four spaces. 339If you have to wrap a long statement, put the operator at the end of the 340line. 341.Bd -literal 342 while (cnt < 20 && this_variable_name_is_too_long_for_its_own_good && 343 ep != NULL) 344 z = a + really + long + statement + that + needs + 345 two lines + gets + indented + four + spaces + 346 on + the + second + and + subsequent + lines; 347.Ed 348.Pp 349Do not add whitespace at the end of a line, and only use tabs 350followed by spaces 351to form the indentation. Do not use more spaces than a tab will produce 352and do not use spaces in front of tabs. 353.Pp 354Closing and opening braces go on the same line as the else. 355Braces that aren't necessary may be left out. 356.Bd -literal 357 if (test) 358 stmt; 359 else if (bar) { 360 stmt; 361 stmt; 362 } else 363 stmt; 364.Ed 365.Pp 366No spaces after function names. Commas have a space after them. No spaces 367after 368.Sq \&( 369or 370.Sq \&[ 371or preceding 372.Sq \&] 373or 374.Sq \&) 375characters. 376.Bd -literal 377 error = function(a1, a2); 378 if (error != 0) 379 exit(error); 380.Ed 381.Pp 382Unary operators don't require spaces, binary operators do. Don't 383use parentheses unless they're required for precedence or unless the 384statement is confusing without them. Remember that other people may 385confuse easier than you. Do YOU understand the following? 386.Bd -literal 387 a = b->c[0] + ~d == (e || f) || g && h ? i : j >> 1; 388 k = !(l & FLAGS); 389.Ed 390.Pp 391Exits should be 0 on success, or according to the predefined 392values in 393.Xr sysexits 3 . 394.Bd -literal 395 exit(EX_OK); /* 396 * Avoid obvious comments such as 397 * "Exit 0 on success." 398 */ 399} 400.Ed 401.Pp 402The function type should be on a line by itself 403preceding the function. 404.Bd -literal 405static char * 406function(int a1, int a2, float fl, int a4) 407{ 408.Ed 409.Pp 410When declaring variables in functions declare them sorted by size, 411then in alphabetical order; multiple ones per line are okay. 412If a line overflows reuse the type keyword. 413.Pp 414Be careful to not obfuscate the code by initializing variables in 415the declarations. Use this feature only thoughtfully. 416DO NOT use function calls in initializers. 417.Bd -literal 418 struct foo one, *two; 419 double three; 420 int *four, five; 421 char *six, seven, eight, nine, ten, eleven, twelve; 422 423 four = myfunction(); 424.Ed 425.Pp 426Do not declare functions inside other functions; ANSI C says that 427such declarations have file scope regardless of the nesting of the 428declaration. Hiding file declarations in what appears to be a local 429scope is undesirable and will elicit complaints from a good compiler. 430.Pp 431Casts and sizeof's are not followed by a space. Note that 432.Xr indent 1 433does not understand this rule. 434.Pp 435NULL is the preferred null pointer constant. Use NULL instead of 436(type *)0 or (type *)NULL in contexts where the compiler knows the 437type, e.g., in assignments. Use (type *)NULL in other contexts, 438in particular for all function args. (Casting is essential for 439variadic args and is necessary for other args if the function prototype 440might not be in scope.) 441Test pointers 442against NULL, e.g., use: 443.Bd -literal 444(p = f()) == NULL 445.Ed 446.Pp 447not: 448.Bd -literal 449!(p = f()) 450.Ed 451.Pp 452Don't use '!' for tests unless it's a boolean, e.g. use 453.Bd -literal 454if (*p == '\e0') 455.Ed 456.Pp 457not 458.Bd -literal 459if (!*p) 460.Ed 461.Pp 462Routines returning void * should not have their return values cast 463to any pointer type. 464.Pp 465Use 466.Xr err 3 467or 468.Xr warn 3 , 469don't roll your own. 470.Bd -literal 471 if ((four = malloc(sizeof(struct foo))) == NULL) 472 err(1, (char *)NULL); 473 if ((six = (int *)overflow()) == NULL) 474 errx(1, "number overflowed"); 475 return (eight); 476} 477.Ed 478.Pp 479Old-style function declarations look like this: 480.Bd -literal 481static char * 482function(a1, a2, fl, a4) 483 int a1, a2; /* Declare ints, too, don't default them. */ 484 float fl; /* Beware double vs. float prototype differences. */ 485 int a4; /* List in order declared. */ 486{ 487.Ed 488.Pp 489Use ANSI function declarations unless you explicitly need K&R compatibility. 490Long parameter lists are wrapped with a normal four space indent. 491.Pp 492Variable numbers of arguments should look like this. 493.Bd -literal 494#include <stdarg.h> 495 496void 497vaf(const char *fmt, ...) 498{ 499 va_list ap; 500 501 va_start(ap, fmt); 502 STUFF; 503 va_end(ap); 504 /* No return needed for void functions. */ 505} 506 507static void 508usage() 509{ 510 /* Insert an empty line if the function has no local variables. */ 511.Ed 512.Pp 513Use 514.Xr printf 3 , 515not fputs/puts/putchar/whatever; it's faster and usually cleaner, not 516to mention avoiding stupid bugs. 517.Pp 518Usage statements should look like the manual pages synopsis. 519The usage statement should be structured in the following order: 520.Pp 521.Bl -enum 522.It 523Options without operands come first, 524in alphabetical order, 525inside a single set of brackets 526.Sq ( \&[ 527and 528.Sq \&] ) . 529.It 530Options with operands come next, 531also in alphabetical order, 532with each option and its argument inside its own pair of brackets. 533.It 534Required arguments 535(if any) 536are next, 537listed in the order they should be specified in the command line. 538.It 539Finally, 540any optional arguments should be listed, 541listed in the order they should be specified, 542and all inside brackets. 543.El 544.Pp 545A bar 546.Pq Sq \&| 547separates either-or options/arguments, 548and multiple options/arguments which are specified together are 549placed in a single set of brackets. 550.Pp 551.Bd -ragged -offset 4n 552"usage: f [-aDde] [-b b_arg] [-m m_arg] req1 req2 [opt1 [opt2]]\en" 553"usage: f [-a | -b] [-c [-dEe] [-n number]]\en" 554.Ed 555.Bd -literal 556 (void)fprintf(stderr, "usage: f [-ab]\en"); 557 exit(EX_USAGE); 558} 559.Ed 560.Pp 561Note that the manual page options description should list the options in 562pure alphabetical order. 563That is, without regard to whether an option takes arguments or not. 564The alphabetical ordering should take into account the case ordering 565shown above. 566.Pp 567New core kernel code should be reasonably compliant with the style guides. 568The guidelines for third-party maintained modules and device drivers are more 569relaxed but at a minimum should be internally consistent with their style. 570.Pp 571Stylistic changes (including whitespace changes) are hard on the source 572repository and are to be avoided without good reason. Code that is 573approximately 574.Fx 575KNF 576.Xr style 9 577compliant in the repository must not diverge from compliance. 578.Pp 579Whenever possible, code should be run through a code checker 580(e.g., 581.Xr lint 1 582or 583"gcc -Wall") and produce minimal warnings. 584.Sh SEE ALSO 585.Xr indent 1 , 586.Xr lint 1 , 587.Xr err 3 , 588.Xr sysexits 3 , 589.Xr warn 3 590.Sh HISTORY 591This man page is largely based on the src/admin/style/style file from 592the 593.Bx 4.4 Lite2 594release, with updates to reflect the current practice and 595desire of the 596.Fx 597project. 598