https://github.com/torvalds/linux
Revision 215ddd6664ced067afca7eebd2d1eb83f064ff5a authored by Mel Gorman on 08 July 2011, 22:39:40 UTC, committed by Linus Torvalds on 09 July 2011, 04:14:43 UTC
During allocator-intensive workloads, kswapd will be woken frequently
causing free memory to oscillate between the high and min watermark.  This
is expected behaviour.  Unfortunately, if the highest zone is small, a
problem occurs.

When balance_pgdat() returns, it may be at a lower classzone_idx than it
started because the highest zone was unreclaimable.  Before checking if it
should go to sleep though, it checks pgdat->classzone_idx which when there
is no other activity will be MAX_NR_ZONES-1.  It interprets this as it has
been woken up while reclaiming, skips scheduling and reclaims again.  As
there is no useful reclaim work to do, it enters into a loop of shrinking
slab consuming loads of CPU until the highest zone becomes reclaimable for
a long period of time.

There are two problems here.  1) If the returned classzone or order is
lower, it'll continue reclaiming without scheduling.  2) if the highest
zone was marked unreclaimable but balance_pgdat() returns immediately at
DEF_PRIORITY, the new lower classzone is not communicated back to kswapd()
for sleeping.

This patch does two things that are related.  If the end_zone is
unreclaimable, this information is communicated back.  Second, if the
classzone or order was reduced due to failing to reclaim, new information
is not read from pgdat and instead an attempt is made to go to sleep.  Due
to this, it is also necessary that pgdat->classzone_idx be initialised
each time to pgdat->nr_zones - 1 to avoid re-reads being interpreted as
wakeups.

Signed-off-by: Mel Gorman <mgorman@suse.de>
Reported-by: Pádraig Brady <P@draigBrady.com>
Tested-by: Pádraig Brady <P@draigBrady.com>
Tested-by: Andrew Lutomirski <luto@mit.edu>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
1 parent da175d0
Raw File
Tip revision: 215ddd6664ced067afca7eebd2d1eb83f064ff5a authored by Mel Gorman on 08 July 2011, 22:39:40 UTC
mm: vmscan: only read new_classzone_idx from pgdat when reclaiming successfully
Tip revision: 215ddd6
unifdef.c
/*
 * Copyright (c) 2002 - 2011 Tony Finch <dot@dotat.at>
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 *
 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE.
 */

/*
 * unifdef - remove ifdef'ed lines
 *
 * This code was derived from software contributed to Berkeley by Dave Yost.
 * It was rewritten to support ANSI C by Tony Finch. The original version
 * of unifdef carried the 4-clause BSD copyright licence. None of its code
 * remains in this version (though some of the names remain) so it now
 * carries a more liberal licence.
 *
 *  Wishlist:
 *      provide an option which will append the name of the
 *        appropriate symbol after #else's and #endif's
 *      provide an option which will check symbols after
 *        #else's and #endif's to see that they match their
 *        corresponding #ifdef or #ifndef
 *
 *   These require better buffer handling, which would also make
 *   it possible to handle all "dodgy" directives correctly.
 */

#include <sys/types.h>
#include <sys/stat.h>

#include <ctype.h>
#include <err.h>
#include <errno.h>
#include <stdarg.h>
#include <stdbool.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>

const char copyright[] =
    "@(#) $Version: unifdef-2.5 $\n"
    "@(#) $Author: Tony Finch (dot@dotat.at) $\n"
    "@(#) $URL: http://dotat.at/prog/unifdef $\n"
;

/* types of input lines: */
typedef enum {
	LT_TRUEI,		/* a true #if with ignore flag */
	LT_FALSEI,		/* a false #if with ignore flag */
	LT_IF,			/* an unknown #if */
	LT_TRUE,		/* a true #if */
	LT_FALSE,		/* a false #if */
	LT_ELIF,		/* an unknown #elif */
	LT_ELTRUE,		/* a true #elif */
	LT_ELFALSE,		/* a false #elif */
	LT_ELSE,		/* #else */
	LT_ENDIF,		/* #endif */
	LT_DODGY,		/* flag: directive is not on one line */
	LT_DODGY_LAST = LT_DODGY + LT_ENDIF,
	LT_PLAIN,		/* ordinary line */
	LT_EOF,			/* end of file */
	LT_ERROR,		/* unevaluable #if */
	LT_COUNT
} Linetype;

static char const * const linetype_name[] = {
	"TRUEI", "FALSEI", "IF", "TRUE", "FALSE",
	"ELIF", "ELTRUE", "ELFALSE", "ELSE", "ENDIF",
	"DODGY TRUEI", "DODGY FALSEI",
	"DODGY IF", "DODGY TRUE", "DODGY FALSE",
	"DODGY ELIF", "DODGY ELTRUE", "DODGY ELFALSE",
	"DODGY ELSE", "DODGY ENDIF",
	"PLAIN", "EOF", "ERROR"
};

/* state of #if processing */
typedef enum {
	IS_OUTSIDE,
	IS_FALSE_PREFIX,	/* false #if followed by false #elifs */
	IS_TRUE_PREFIX,		/* first non-false #(el)if is true */
	IS_PASS_MIDDLE,		/* first non-false #(el)if is unknown */
	IS_FALSE_MIDDLE,	/* a false #elif after a pass state */
	IS_TRUE_MIDDLE,		/* a true #elif after a pass state */
	IS_PASS_ELSE,		/* an else after a pass state */
	IS_FALSE_ELSE,		/* an else after a true state */
	IS_TRUE_ELSE,		/* an else after only false states */
	IS_FALSE_TRAILER,	/* #elifs after a true are false */
	IS_COUNT
} Ifstate;

static char const * const ifstate_name[] = {
	"OUTSIDE", "FALSE_PREFIX", "TRUE_PREFIX",
	"PASS_MIDDLE", "FALSE_MIDDLE", "TRUE_MIDDLE",
	"PASS_ELSE", "FALSE_ELSE", "TRUE_ELSE",
	"FALSE_TRAILER"
};

/* state of comment parser */
typedef enum {
	NO_COMMENT = false,	/* outside a comment */
	C_COMMENT,		/* in a comment like this one */
	CXX_COMMENT,		/* between // and end of line */
	STARTING_COMMENT,	/* just after slash-backslash-newline */
	FINISHING_COMMENT,	/* star-backslash-newline in a C comment */
	CHAR_LITERAL,		/* inside '' */
	STRING_LITERAL		/* inside "" */
} Comment_state;

static char const * const comment_name[] = {
	"NO", "C", "CXX", "STARTING", "FINISHING", "CHAR", "STRING"
};

/* state of preprocessor line parser */
typedef enum {
	LS_START,		/* only space and comments on this line */
	LS_HASH,		/* only space, comments, and a hash */
	LS_DIRTY		/* this line can't be a preprocessor line */
} Line_state;

static char const * const linestate_name[] = {
	"START", "HASH", "DIRTY"
};

/*
 * Minimum translation limits from ISO/IEC 9899:1999 5.2.4.1
 */
#define	MAXDEPTH        64			/* maximum #if nesting */
#define	MAXLINE         4096			/* maximum length of line */
#define	MAXSYMS         4096			/* maximum number of symbols */

/*
 * Sometimes when editing a keyword the replacement text is longer, so
 * we leave some space at the end of the tline buffer to accommodate this.
 */
#define	EDITSLOP        10

/*
 * For temporary filenames
 */
#define TEMPLATE        "unifdef.XXXXXX"

/*
 * Globals.
 */

static bool             compblank;		/* -B: compress blank lines */
static bool             lnblank;		/* -b: blank deleted lines */
static bool             complement;		/* -c: do the complement */
static bool             debugging;		/* -d: debugging reports */
static bool             iocccok;		/* -e: fewer IOCCC errors */
static bool             strictlogic;		/* -K: keep ambiguous #ifs */
static bool             killconsts;		/* -k: eval constant #ifs */
static bool             lnnum;			/* -n: add #line directives */
static bool             symlist;		/* -s: output symbol list */
static bool             symdepth;		/* -S: output symbol depth */
static bool             text;			/* -t: this is a text file */

static const char      *symname[MAXSYMS];	/* symbol name */
static const char      *value[MAXSYMS];		/* -Dsym=value */
static bool             ignore[MAXSYMS];	/* -iDsym or -iUsym */
static int              nsyms;			/* number of symbols */

static FILE            *input;			/* input file pointer */
static const char      *filename;		/* input file name */
static int              linenum;		/* current line number */
static FILE            *output;			/* output file pointer */
static const char      *ofilename;		/* output file name */
static bool             overwriting;		/* output overwrites input */
static char             tempname[FILENAME_MAX];	/* used when overwriting */

static char             tline[MAXLINE+EDITSLOP];/* input buffer plus space */
static char            *keyword;		/* used for editing #elif's */

static const char      *newline;		/* input file format */
static const char       newline_unix[] = "\n";
static const char       newline_crlf[] = "\r\n";

static Comment_state    incomment;		/* comment parser state */
static Line_state       linestate;		/* #if line parser state */
static Ifstate          ifstate[MAXDEPTH];	/* #if processor state */
static bool             ignoring[MAXDEPTH];	/* ignore comments state */
static int              stifline[MAXDEPTH];	/* start of current #if */
static int              depth;			/* current #if nesting */
static int              delcount;		/* count of deleted lines */
static unsigned         blankcount;		/* count of blank lines */
static unsigned         blankmax;		/* maximum recent blankcount */
static bool             constexpr;		/* constant #if expression */
static bool             zerosyms = true;	/* to format symdepth output */
static bool             firstsym;		/* ditto */

static int              exitstat;		/* program exit status */

static void             addsym(bool, bool, char *);
static void             closeout(void);
static void             debug(const char *, ...);
static void             done(void);
static void             error(const char *);
static int              findsym(const char *);
static void             flushline(bool);
static Linetype         parseline(void);
static Linetype         ifeval(const char **);
static void             ignoreoff(void);
static void             ignoreon(void);
static void             keywordedit(const char *);
static void             nest(void);
static void             process(void);
static const char      *skipargs(const char *);
static const char      *skipcomment(const char *);
static const char      *skipsym(const char *);
static void             state(Ifstate);
static int              strlcmp(const char *, const char *, size_t);
static void             unnest(void);
static void             usage(void);
static void             version(void);

#define endsym(c) (!isalnum((unsigned char)c) && c != '_')

/*
 * The main program.
 */
int
main(int argc, char *argv[])
{
	int opt;

	while ((opt = getopt(argc, argv, "i:D:U:I:o:bBcdeKklnsStV")) != -1)
		switch (opt) {
		case 'i': /* treat stuff controlled by these symbols as text */
			/*
			 * For strict backwards-compatibility the U or D
			 * should be immediately after the -i but it doesn't
			 * matter much if we relax that requirement.
			 */
			opt = *optarg++;
			if (opt == 'D')
				addsym(true, true, optarg);
			else if (opt == 'U')
				addsym(true, false, optarg);
			else
				usage();
			break;
		case 'D': /* define a symbol */
			addsym(false, true, optarg);
			break;
		case 'U': /* undef a symbol */
			addsym(false, false, optarg);
			break;
		case 'I': /* no-op for compatibility with cpp */
			break;
		case 'b': /* blank deleted lines instead of omitting them */
		case 'l': /* backwards compatibility */
			lnblank = true;
			break;
		case 'B': /* compress blank lines around removed section */
			compblank = true;
			break;
		case 'c': /* treat -D as -U and vice versa */
			complement = true;
			break;
		case 'd':
			debugging = true;
			break;
		case 'e': /* fewer errors from dodgy lines */
			iocccok = true;
			break;
		case 'K': /* keep ambiguous #ifs */
			strictlogic = true;
			break;
		case 'k': /* process constant #ifs */
			killconsts = true;
			break;
		case 'n': /* add #line directive after deleted lines */
			lnnum = true;
			break;
		case 'o': /* output to a file */
			ofilename = optarg;
			break;
		case 's': /* only output list of symbols that control #ifs */
			symlist = true;
			break;
		case 'S': /* list symbols with their nesting depth */
			symlist = symdepth = true;
			break;
		case 't': /* don't parse C comments */
			text = true;
			break;
		case 'V': /* print version */
			version();
		default:
			usage();
		}
	argc -= optind;
	argv += optind;
	if (compblank && lnblank)
		errx(2, "-B and -b are mutually exclusive");
	if (argc > 1) {
		errx(2, "can only do one file");
	} else if (argc == 1 && strcmp(*argv, "-") != 0) {
		filename = *argv;
		input = fopen(filename, "rb");
		if (input == NULL)
			err(2, "can't open %s", filename);
	} else {
		filename = "[stdin]";
		input = stdin;
	}
	if (ofilename == NULL) {
		ofilename = "[stdout]";
		output = stdout;
	} else {
		struct stat ist, ost;
		if (stat(ofilename, &ost) == 0 &&
		    fstat(fileno(input), &ist) == 0)
			overwriting = (ist.st_dev == ost.st_dev
				    && ist.st_ino == ost.st_ino);
		if (overwriting) {
			const char *dirsep;
			int ofd;

			dirsep = strrchr(ofilename, '/');
			if (dirsep != NULL)
				snprintf(tempname, sizeof(tempname),
				    "%.*s/" TEMPLATE,
				    (int)(dirsep - ofilename), ofilename);
			else
				snprintf(tempname, sizeof(tempname),
				    TEMPLATE);
			ofd = mkstemp(tempname);
			if (ofd != -1)
				output = fdopen(ofd, "wb+");
			if (output == NULL)
				err(2, "can't create temporary file");
			fchmod(ofd, ist.st_mode & (S_IRWXU|S_IRWXG|S_IRWXO));
		} else {
			output = fopen(ofilename, "wb");
			if (output == NULL)
				err(2, "can't open %s", ofilename);
		}
	}
	process();
	abort(); /* bug */
}

static void
version(void)
{
	const char *c = copyright;
	for (;;) {
		while (*++c != '$')
			if (*c == '\0')
				exit(0);
		while (*++c != '$')
			putc(*c, stderr);
		putc('\n', stderr);
	}
}

static void
usage(void)
{
	fprintf(stderr, "usage: unifdef [-bBcdeKknsStV] [-Ipath]"
	    " [-Dsym[=val]] [-Usym] [-iDsym[=val]] [-iUsym] ... [file]\n");
	exit(2);
}

/*
 * A state transition function alters the global #if processing state
 * in a particular way. The table below is indexed by the current
 * processing state and the type of the current line.
 *
 * Nesting is handled by keeping a stack of states; some transition
 * functions increase or decrease the depth. They also maintain the
 * ignore state on a stack. In some complicated cases they have to
 * alter the preprocessor directive, as follows.
 *
 * When we have processed a group that starts off with a known-false
 * #if/#elif sequence (which has therefore been deleted) followed by a
 * #elif that we don't understand and therefore must keep, we edit the
 * latter into a #if to keep the nesting correct. We use strncpy() to
 * overwrite the 4 byte token "elif" with "if  " without a '\0' byte.
 *
 * When we find a true #elif in a group, the following block will
 * always be kept and the rest of the sequence after the next #elif or
 * #else will be discarded. We edit the #elif into a #else and the
 * following directive to #endif since this has the desired behaviour.
 *
 * "Dodgy" directives are split across multiple lines, the most common
 * example being a multi-line comment hanging off the right of the
 * directive. We can handle them correctly only if there is no change
 * from printing to dropping (or vice versa) caused by that directive.
 * If the directive is the first of a group we have a choice between
 * failing with an error, or passing it through unchanged instead of
 * evaluating it. The latter is not the default to avoid questions from
 * users about unifdef unexpectedly leaving behind preprocessor directives.
 */
typedef void state_fn(void);

/* report an error */
static void Eelif (void) { error("Inappropriate #elif"); }
static void Eelse (void) { error("Inappropriate #else"); }
static void Eendif(void) { error("Inappropriate #endif"); }
static void Eeof  (void) { error("Premature EOF"); }
static void Eioccc(void) { error("Obfuscated preprocessor control line"); }
/* plain line handling */
static void print (void) { flushline(true); }
static void drop  (void) { flushline(false); }
/* output lacks group's start line */
static void Strue (void) { drop();  ignoreoff(); state(IS_TRUE_PREFIX); }
static void Sfalse(void) { drop();  ignoreoff(); state(IS_FALSE_PREFIX); }
static void Selse (void) { drop();               state(IS_TRUE_ELSE); }
/* print/pass this block */
static void Pelif (void) { print(); ignoreoff(); state(IS_PASS_MIDDLE); }
static void Pelse (void) { print();              state(IS_PASS_ELSE); }
static void Pendif(void) { print(); unnest(); }
/* discard this block */
static void Dfalse(void) { drop();  ignoreoff(); state(IS_FALSE_TRAILER); }
static void Delif (void) { drop();  ignoreoff(); state(IS_FALSE_MIDDLE); }
static void Delse (void) { drop();               state(IS_FALSE_ELSE); }
static void Dendif(void) { drop();  unnest(); }
/* first line of group */
static void Fdrop (void) { nest();  Dfalse(); }
static void Fpass (void) { nest();  Pelif(); }
static void Ftrue (void) { nest();  Strue(); }
static void Ffalse(void) { nest();  Sfalse(); }
/* variable pedantry for obfuscated lines */
static void Oiffy (void) { if (!iocccok) Eioccc(); Fpass(); ignoreon(); }
static void Oif   (void) { if (!iocccok) Eioccc(); Fpass(); }
static void Oelif (void) { if (!iocccok) Eioccc(); Pelif(); }
/* ignore comments in this block */
static void Idrop (void) { Fdrop();  ignoreon(); }
static void Itrue (void) { Ftrue();  ignoreon(); }
static void Ifalse(void) { Ffalse(); ignoreon(); }
/* modify this line */
static void Mpass (void) { strncpy(keyword, "if  ", 4); Pelif(); }
static void Mtrue (void) { keywordedit("else");  state(IS_TRUE_MIDDLE); }
static void Melif (void) { keywordedit("endif"); state(IS_FALSE_TRAILER); }
static void Melse (void) { keywordedit("endif"); state(IS_FALSE_ELSE); }

static state_fn * const trans_table[IS_COUNT][LT_COUNT] = {
/* IS_OUTSIDE */
{ Itrue, Ifalse,Fpass, Ftrue, Ffalse,Eelif, Eelif, Eelif, Eelse, Eendif,
  Oiffy, Oiffy, Fpass, Oif,   Oif,   Eelif, Eelif, Eelif, Eelse, Eendif,
  print, done,  abort },
/* IS_FALSE_PREFIX */
{ Idrop, Idrop, Fdrop, Fdrop, Fdrop, Mpass, Strue, Sfalse,Selse, Dendif,
  Idrop, Idrop, Fdrop, Fdrop, Fdrop, Mpass, Eioccc,Eioccc,Eioccc,Eioccc,
  drop,  Eeof,  abort },
/* IS_TRUE_PREFIX */
{ Itrue, Ifalse,Fpass, Ftrue, Ffalse,Dfalse,Dfalse,Dfalse,Delse, Dendif,
  Oiffy, Oiffy, Fpass, Oif,   Oif,   Eioccc,Eioccc,Eioccc,Eioccc,Eioccc,
  print, Eeof,  abort },
/* IS_PASS_MIDDLE */
{ Itrue, Ifalse,Fpass, Ftrue, Ffalse,Pelif, Mtrue, Delif, Pelse, Pendif,
  Oiffy, Oiffy, Fpass, Oif,   Oif,   Pelif, Oelif, Oelif, Pelse, Pendif,
  print, Eeof,  abort },
/* IS_FALSE_MIDDLE */
{ Idrop, Idrop, Fdrop, Fdrop, Fdrop, Pelif, Mtrue, Delif, Pelse, Pendif,
  Idrop, Idrop, Fdrop, Fdrop, Fdrop, Eioccc,Eioccc,Eioccc,Eioccc,Eioccc,
  drop,  Eeof,  abort },
/* IS_TRUE_MIDDLE */
{ Itrue, Ifalse,Fpass, Ftrue, Ffalse,Melif, Melif, Melif, Melse, Pendif,
  Oiffy, Oiffy, Fpass, Oif,   Oif,   Eioccc,Eioccc,Eioccc,Eioccc,Pendif,
  print, Eeof,  abort },
/* IS_PASS_ELSE */
{ Itrue, Ifalse,Fpass, Ftrue, Ffalse,Eelif, Eelif, Eelif, Eelse, Pendif,
  Oiffy, Oiffy, Fpass, Oif,   Oif,   Eelif, Eelif, Eelif, Eelse, Pendif,
  print, Eeof,  abort },
/* IS_FALSE_ELSE */
{ Idrop, Idrop, Fdrop, Fdrop, Fdrop, Eelif, Eelif, Eelif, Eelse, Dendif,
  Idrop, Idrop, Fdrop, Fdrop, Fdrop, Eelif, Eelif, Eelif, Eelse, Eioccc,
  drop,  Eeof,  abort },
/* IS_TRUE_ELSE */
{ Itrue, Ifalse,Fpass, Ftrue, Ffalse,Eelif, Eelif, Eelif, Eelse, Dendif,
  Oiffy, Oiffy, Fpass, Oif,   Oif,   Eelif, Eelif, Eelif, Eelse, Eioccc,
  print, Eeof,  abort },
/* IS_FALSE_TRAILER */
{ Idrop, Idrop, Fdrop, Fdrop, Fdrop, Dfalse,Dfalse,Dfalse,Delse, Dendif,
  Idrop, Idrop, Fdrop, Fdrop, Fdrop, Dfalse,Dfalse,Dfalse,Delse, Eioccc,
  drop,  Eeof,  abort }
/*TRUEI  FALSEI IF     TRUE   FALSE  ELIF   ELTRUE ELFALSE ELSE  ENDIF
  TRUEI  FALSEI IF     TRUE   FALSE  ELIF   ELTRUE ELFALSE ELSE  ENDIF (DODGY)
  PLAIN  EOF    ERROR */
};

/*
 * State machine utility functions
 */
static void
ignoreoff(void)
{
	if (depth == 0)
		abort(); /* bug */
	ignoring[depth] = ignoring[depth-1];
}
static void
ignoreon(void)
{
	ignoring[depth] = true;
}
static void
keywordedit(const char *replacement)
{
	snprintf(keyword, tline + sizeof(tline) - keyword,
	    "%s%s", replacement, newline);
	print();
}
static void
nest(void)
{
	if (depth > MAXDEPTH-1)
		abort(); /* bug */
	if (depth == MAXDEPTH-1)
		error("Too many levels of nesting");
	depth += 1;
	stifline[depth] = linenum;
}
static void
unnest(void)
{
	if (depth == 0)
		abort(); /* bug */
	depth -= 1;
}
static void
state(Ifstate is)
{
	ifstate[depth] = is;
}

/*
 * Write a line to the output or not, according to command line options.
 */
static void
flushline(bool keep)
{
	if (symlist)
		return;
	if (keep ^ complement) {
		bool blankline = tline[strspn(tline, " \t\r\n")] == '\0';
		if (blankline && compblank && blankcount != blankmax) {
			delcount += 1;
			blankcount += 1;
		} else {
			if (lnnum && delcount > 0)
				printf("#line %d%s", linenum, newline);
			fputs(tline, output);
			delcount = 0;
			blankmax = blankcount = blankline ? blankcount + 1 : 0;
		}
	} else {
		if (lnblank)
			fputs(newline, output);
		exitstat = 1;
		delcount += 1;
		blankcount = 0;
	}
	if (debugging)
		fflush(output);
}

/*
 * The driver for the state machine.
 */
static void
process(void)
{
	/* When compressing blank lines, act as if the file
	   is preceded by a large number of blank lines. */
	blankmax = blankcount = 1000;
	for (;;) {
		Linetype lineval = parseline();
		trans_table[ifstate[depth]][lineval]();
		debug("process line %d %s -> %s depth %d",
		    linenum, linetype_name[lineval],
		    ifstate_name[ifstate[depth]], depth);
	}
}

/*
 * Flush the output and handle errors.
 */
static void
closeout(void)
{
	if (symdepth && !zerosyms)
		printf("\n");
	if (fclose(output) == EOF) {
		warn("couldn't write to %s", ofilename);
		if (overwriting) {
			unlink(tempname);
			errx(2, "%s unchanged", filename);
		} else {
			exit(2);
		}
	}
}

/*
 * Clean up and exit.
 */
static void
done(void)
{
	if (incomment)
		error("EOF in comment");
	closeout();
	if (overwriting && rename(tempname, ofilename) == -1) {
		warn("couldn't rename temporary file");
		unlink(tempname);
		errx(2, "%s unchanged", ofilename);
	}
	exit(exitstat);
}

/*
 * Parse a line and determine its type. We keep the preprocessor line
 * parser state between calls in the global variable linestate, with
 * help from skipcomment().
 */
static Linetype
parseline(void)
{
	const char *cp;
	int cursym;
	int kwlen;
	Linetype retval;
	Comment_state wascomment;

	linenum++;
	if (fgets(tline, MAXLINE, input) == NULL)
		return (LT_EOF);
	if (newline == NULL) {
		if (strrchr(tline, '\n') == strrchr(tline, '\r') + 1)
			newline = newline_crlf;
		else
			newline = newline_unix;
	}
	retval = LT_PLAIN;
	wascomment = incomment;
	cp = skipcomment(tline);
	if (linestate == LS_START) {
		if (*cp == '#') {
			linestate = LS_HASH;
			firstsym = true;
			cp = skipcomment(cp + 1);
		} else if (*cp != '\0')
			linestate = LS_DIRTY;
	}
	if (!incomment && linestate == LS_HASH) {
		keyword = tline + (cp - tline);
		cp = skipsym(cp);
		kwlen = cp - keyword;
		/* no way can we deal with a continuation inside a keyword */
		if (strncmp(cp, "\\\r\n", 3) == 0 ||
		    strncmp(cp, "\\\n", 2) == 0)
			Eioccc();
		if (strlcmp("ifdef", keyword, kwlen) == 0 ||
		    strlcmp("ifndef", keyword, kwlen) == 0) {
			cp = skipcomment(cp);
			if ((cursym = findsym(cp)) < 0)
				retval = LT_IF;
			else {
				retval = (keyword[2] == 'n')
				    ? LT_FALSE : LT_TRUE;
				if (value[cursym] == NULL)
					retval = (retval == LT_TRUE)
					    ? LT_FALSE : LT_TRUE;
				if (ignore[cursym])
					retval = (retval == LT_TRUE)
					    ? LT_TRUEI : LT_FALSEI;
			}
			cp = skipsym(cp);
		} else if (strlcmp("if", keyword, kwlen) == 0)
			retval = ifeval(&cp);
		else if (strlcmp("elif", keyword, kwlen) == 0)
			retval = ifeval(&cp) - LT_IF + LT_ELIF;
		else if (strlcmp("else", keyword, kwlen) == 0)
			retval = LT_ELSE;
		else if (strlcmp("endif", keyword, kwlen) == 0)
			retval = LT_ENDIF;
		else {
			linestate = LS_DIRTY;
			retval = LT_PLAIN;
		}
		cp = skipcomment(cp);
		if (*cp != '\0') {
			linestate = LS_DIRTY;
			if (retval == LT_TRUE || retval == LT_FALSE ||
			    retval == LT_TRUEI || retval == LT_FALSEI)
				retval = LT_IF;
			if (retval == LT_ELTRUE || retval == LT_ELFALSE)
				retval = LT_ELIF;
		}
		if (retval != LT_PLAIN && (wascomment || incomment)) {
			retval += LT_DODGY;
			if (incomment)
				linestate = LS_DIRTY;
		}
		/* skipcomment normally changes the state, except
		   if the last line of the file lacks a newline, or
		   if there is too much whitespace in a directive */
		if (linestate == LS_HASH) {
			size_t len = cp - tline;
			if (fgets(tline + len, MAXLINE - len, input) == NULL) {
				/* append the missing newline */
				strcpy(tline + len, newline);
				cp += strlen(newline);
				linestate = LS_START;
			} else {
				linestate = LS_DIRTY;
			}
		}
	}
	if (linestate == LS_DIRTY) {
		while (*cp != '\0')
			cp = skipcomment(cp + 1);
	}
	debug("parser line %d state %s comment %s line", linenum,
	    comment_name[incomment], linestate_name[linestate]);
	return (retval);
}

/*
 * These are the binary operators that are supported by the expression
 * evaluator.
 */
static Linetype op_strict(int *p, int v, Linetype at, Linetype bt) {
	if(at == LT_IF || bt == LT_IF) return (LT_IF);
	return (*p = v, v ? LT_TRUE : LT_FALSE);
}
static Linetype op_lt(int *p, Linetype at, int a, Linetype bt, int b) {
	return op_strict(p, a < b, at, bt);
}
static Linetype op_gt(int *p, Linetype at, int a, Linetype bt, int b) {
	return op_strict(p, a > b, at, bt);
}
static Linetype op_le(int *p, Linetype at, int a, Linetype bt, int b) {
	return op_strict(p, a <= b, at, bt);
}
static Linetype op_ge(int *p, Linetype at, int a, Linetype bt, int b) {
	return op_strict(p, a >= b, at, bt);
}
static Linetype op_eq(int *p, Linetype at, int a, Linetype bt, int b) {
	return op_strict(p, a == b, at, bt);
}
static Linetype op_ne(int *p, Linetype at, int a, Linetype bt, int b) {
	return op_strict(p, a != b, at, bt);
}
static Linetype op_or(int *p, Linetype at, int a, Linetype bt, int b) {
	if (!strictlogic && (at == LT_TRUE || bt == LT_TRUE))
		return (*p = 1, LT_TRUE);
	return op_strict(p, a || b, at, bt);
}
static Linetype op_and(int *p, Linetype at, int a, Linetype bt, int b) {
	if (!strictlogic && (at == LT_FALSE || bt == LT_FALSE))
		return (*p = 0, LT_FALSE);
	return op_strict(p, a && b, at, bt);
}

/*
 * An evaluation function takes three arguments, as follows: (1) a pointer to
 * an element of the precedence table which lists the operators at the current
 * level of precedence; (2) a pointer to an integer which will receive the
 * value of the expression; and (3) a pointer to a char* that points to the
 * expression to be evaluated and that is updated to the end of the expression
 * when evaluation is complete. The function returns LT_FALSE if the value of
 * the expression is zero, LT_TRUE if it is non-zero, LT_IF if the expression
 * depends on an unknown symbol, or LT_ERROR if there is a parse failure.
 */
struct ops;

typedef Linetype eval_fn(const struct ops *, int *, const char **);

static eval_fn eval_table, eval_unary;

/*
 * The precedence table. Expressions involving binary operators are evaluated
 * in a table-driven way by eval_table. When it evaluates a subexpression it
 * calls the inner function with its first argument pointing to the next
 * element of the table. Innermost expressions have special non-table-driven
 * handling.
 */
static const struct ops {
	eval_fn *inner;
	struct op {
		const char *str;
		Linetype (*fn)(int *, Linetype, int, Linetype, int);
	} op[5];
} eval_ops[] = {
	{ eval_table, { { "||", op_or } } },
	{ eval_table, { { "&&", op_and } } },
	{ eval_table, { { "==", op_eq },
			{ "!=", op_ne } } },
	{ eval_unary, { { "<=", op_le },
			{ ">=", op_ge },
			{ "<", op_lt },
			{ ">", op_gt } } }
};

/*
 * Function for evaluating the innermost parts of expressions,
 * viz. !expr (expr) number defined(symbol) symbol
 * We reset the constexpr flag in the last two cases.
 */
static Linetype
eval_unary(const struct ops *ops, int *valp, const char **cpp)
{
	const char *cp;
	char *ep;
	int sym;
	bool defparen;
	Linetype lt;

	cp = skipcomment(*cpp);
	if (*cp == '!') {
		debug("eval%d !", ops - eval_ops);
		cp++;
		lt = eval_unary(ops, valp, &cp);
		if (lt == LT_ERROR)
			return (LT_ERROR);
		if (lt != LT_IF) {
			*valp = !*valp;
			lt = *valp ? LT_TRUE : LT_FALSE;
		}
	} else if (*cp == '(') {
		cp++;
		debug("eval%d (", ops - eval_ops);
		lt = eval_table(eval_ops, valp, &cp);
		if (lt == LT_ERROR)
			return (LT_ERROR);
		cp = skipcomment(cp);
		if (*cp++ != ')')
			return (LT_ERROR);
	} else if (isdigit((unsigned char)*cp)) {
		debug("eval%d number", ops - eval_ops);
		*valp = strtol(cp, &ep, 0);
		if (ep == cp)
			return (LT_ERROR);
		lt = *valp ? LT_TRUE : LT_FALSE;
		cp = skipsym(cp);
	} else if (strncmp(cp, "defined", 7) == 0 && endsym(cp[7])) {
		cp = skipcomment(cp+7);
		debug("eval%d defined", ops - eval_ops);
		if (*cp == '(') {
			cp = skipcomment(cp+1);
			defparen = true;
		} else {
			defparen = false;
		}
		sym = findsym(cp);
		if (sym < 0) {
			lt = LT_IF;
		} else {
			*valp = (value[sym] != NULL);
			lt = *valp ? LT_TRUE : LT_FALSE;
		}
		cp = skipsym(cp);
		cp = skipcomment(cp);
		if (defparen && *cp++ != ')')
			return (LT_ERROR);
		constexpr = false;
	} else if (!endsym(*cp)) {
		debug("eval%d symbol", ops - eval_ops);
		sym = findsym(cp);
		cp = skipsym(cp);
		if (sym < 0) {
			lt = LT_IF;
			cp = skipargs(cp);
		} else if (value[sym] == NULL) {
			*valp = 0;
			lt = LT_FALSE;
		} else {
			*valp = strtol(value[sym], &ep, 0);
			if (*ep != '\0' || ep == value[sym])
				return (LT_ERROR);
			lt = *valp ? LT_TRUE : LT_FALSE;
			cp = skipargs(cp);
		}
		constexpr = false;
	} else {
		debug("eval%d bad expr", ops - eval_ops);
		return (LT_ERROR);
	}

	*cpp = cp;
	debug("eval%d = %d", ops - eval_ops, *valp);
	return (lt);
}

/*
 * Table-driven evaluation of binary operators.
 */
static Linetype
eval_table(const struct ops *ops, int *valp, const char **cpp)
{
	const struct op *op;
	const char *cp;
	int val;
	Linetype lt, rt;

	debug("eval%d", ops - eval_ops);
	cp = *cpp;
	lt = ops->inner(ops+1, valp, &cp);
	if (lt == LT_ERROR)
		return (LT_ERROR);
	for (;;) {
		cp = skipcomment(cp);
		for (op = ops->op; op->str != NULL; op++)
			if (strncmp(cp, op->str, strlen(op->str)) == 0)
				break;
		if (op->str == NULL)
			break;
		cp += strlen(op->str);
		debug("eval%d %s", ops - eval_ops, op->str);
		rt = ops->inner(ops+1, &val, &cp);
		if (rt == LT_ERROR)
			return (LT_ERROR);
		lt = op->fn(valp, lt, *valp, rt, val);
	}

	*cpp = cp;
	debug("eval%d = %d", ops - eval_ops, *valp);
	debug("eval%d lt = %s", ops - eval_ops, linetype_name[lt]);
	return (lt);
}

/*
 * Evaluate the expression on a #if or #elif line. If we can work out
 * the result we return LT_TRUE or LT_FALSE accordingly, otherwise we
 * return just a generic LT_IF.
 */
static Linetype
ifeval(const char **cpp)
{
	int ret;
	int val = 0;

	debug("eval %s", *cpp);
	constexpr = killconsts ? false : true;
	ret = eval_table(eval_ops, &val, cpp);
	debug("eval = %d", val);
	return (constexpr ? LT_IF : ret == LT_ERROR ? LT_IF : ret);
}

/*
 * Skip over comments, strings, and character literals and stop at the
 * next character position that is not whitespace. Between calls we keep
 * the comment state in the global variable incomment, and we also adjust
 * the global variable linestate when we see a newline.
 * XXX: doesn't cope with the buffer splitting inside a state transition.
 */
static const char *
skipcomment(const char *cp)
{
	if (text || ignoring[depth]) {
		for (; isspace((unsigned char)*cp); cp++)
			if (*cp == '\n')
				linestate = LS_START;
		return (cp);
	}
	while (*cp != '\0')
		/* don't reset to LS_START after a line continuation */
		if (strncmp(cp, "\\\r\n", 3) == 0)
			cp += 3;
		else if (strncmp(cp, "\\\n", 2) == 0)
			cp += 2;
		else switch (incomment) {
		case NO_COMMENT:
			if (strncmp(cp, "/\\\r\n", 4) == 0) {
				incomment = STARTING_COMMENT;
				cp += 4;
			} else if (strncmp(cp, "/\\\n", 3) == 0) {
				incomment = STARTING_COMMENT;
				cp += 3;
			} else if (strncmp(cp, "/*", 2) == 0) {
				incomment = C_COMMENT;
				cp += 2;
			} else if (strncmp(cp, "//", 2) == 0) {
				incomment = CXX_COMMENT;
				cp += 2;
			} else if (strncmp(cp, "\'", 1) == 0) {
				incomment = CHAR_LITERAL;
				linestate = LS_DIRTY;
				cp += 1;
			} else if (strncmp(cp, "\"", 1) == 0) {
				incomment = STRING_LITERAL;
				linestate = LS_DIRTY;
				cp += 1;
			} else if (strncmp(cp, "\n", 1) == 0) {
				linestate = LS_START;
				cp += 1;
			} else if (strchr(" \r\t", *cp) != NULL) {
				cp += 1;
			} else
				return (cp);
			continue;
		case CXX_COMMENT:
			if (strncmp(cp, "\n", 1) == 0) {
				incomment = NO_COMMENT;
				linestate = LS_START;
			}
			cp += 1;
			continue;
		case CHAR_LITERAL:
		case STRING_LITERAL:
			if ((incomment == CHAR_LITERAL && cp[0] == '\'') ||
			    (incomment == STRING_LITERAL && cp[0] == '\"')) {
				incomment = NO_COMMENT;
				cp += 1;
			} else if (cp[0] == '\\') {
				if (cp[1] == '\0')
					cp += 1;
				else
					cp += 2;
			} else if (strncmp(cp, "\n", 1) == 0) {
				if (incomment == CHAR_LITERAL)
					error("unterminated char literal");
				else
					error("unterminated string literal");
			} else
				cp += 1;
			continue;
		case C_COMMENT:
			if (strncmp(cp, "*\\\r\n", 4) == 0) {
				incomment = FINISHING_COMMENT;
				cp += 4;
			} else if (strncmp(cp, "*\\\n", 3) == 0) {
				incomment = FINISHING_COMMENT;
				cp += 3;
			} else if (strncmp(cp, "*/", 2) == 0) {
				incomment = NO_COMMENT;
				cp += 2;
			} else
				cp += 1;
			continue;
		case STARTING_COMMENT:
			if (*cp == '*') {
				incomment = C_COMMENT;
				cp += 1;
			} else if (*cp == '/') {
				incomment = CXX_COMMENT;
				cp += 1;
			} else {
				incomment = NO_COMMENT;
				linestate = LS_DIRTY;
			}
			continue;
		case FINISHING_COMMENT:
			if (*cp == '/') {
				incomment = NO_COMMENT;
				cp += 1;
			} else
				incomment = C_COMMENT;
			continue;
		default:
			abort(); /* bug */
		}
	return (cp);
}

/*
 * Skip macro arguments.
 */
static const char *
skipargs(const char *cp)
{
	const char *ocp = cp;
	int level = 0;
	cp = skipcomment(cp);
	if (*cp != '(')
		return (cp);
	do {
		if (*cp == '(')
			level++;
		if (*cp == ')')
			level--;
		cp = skipcomment(cp+1);
	} while (level != 0 && *cp != '\0');
	if (level == 0)
		return (cp);
	else
	/* Rewind and re-detect the syntax error later. */
		return (ocp);
}

/*
 * Skip over an identifier.
 */
static const char *
skipsym(const char *cp)
{
	while (!endsym(*cp))
		++cp;
	return (cp);
}

/*
 * Look for the symbol in the symbol table. If it is found, we return
 * the symbol table index, else we return -1.
 */
static int
findsym(const char *str)
{
	const char *cp;
	int symind;

	cp = skipsym(str);
	if (cp == str)
		return (-1);
	if (symlist) {
		if (symdepth && firstsym)
			printf("%s%3d", zerosyms ? "" : "\n", depth);
		firstsym = zerosyms = false;
		printf("%s%.*s%s",
		    symdepth ? " " : "",
		    (int)(cp-str), str,
		    symdepth ? "" : "\n");
		/* we don't care about the value of the symbol */
		return (0);
	}
	for (symind = 0; symind < nsyms; ++symind) {
		if (strlcmp(symname[symind], str, cp-str) == 0) {
			debug("findsym %s %s", symname[symind],
			    value[symind] ? value[symind] : "");
			return (symind);
		}
	}
	return (-1);
}

/*
 * Add a symbol to the symbol table.
 */
static void
addsym(bool ignorethis, bool definethis, char *sym)
{
	int symind;
	char *val;

	symind = findsym(sym);
	if (symind < 0) {
		if (nsyms >= MAXSYMS)
			errx(2, "too many symbols");
		symind = nsyms++;
	}
	symname[symind] = sym;
	ignore[symind] = ignorethis;
	val = sym + (skipsym(sym) - sym);
	if (definethis) {
		if (*val == '=') {
			value[symind] = val+1;
			*val = '\0';
		} else if (*val == '\0')
			value[symind] = "1";
		else
			usage();
	} else {
		if (*val != '\0')
			usage();
		value[symind] = NULL;
	}
	debug("addsym %s=%s", symname[symind],
	    value[symind] ? value[symind] : "undef");
}

/*
 * Compare s with n characters of t.
 * The same as strncmp() except that it checks that s[n] == '\0'.
 */
static int
strlcmp(const char *s, const char *t, size_t n)
{
	while (n-- && *t != '\0')
		if (*s != *t)
			return ((unsigned char)*s - (unsigned char)*t);
		else
			++s, ++t;
	return ((unsigned char)*s);
}

/*
 * Diagnostics.
 */
static void
debug(const char *msg, ...)
{
	va_list ap;

	if (debugging) {
		va_start(ap, msg);
		vwarnx(msg, ap);
		va_end(ap);
	}
}

static void
error(const char *msg)
{
	if (depth == 0)
		warnx("%s: %d: %s", filename, linenum, msg);
	else
		warnx("%s: %d: %s (#if line %d depth %d)",
		    filename, linenum, msg, stifline[depth], depth);
	closeout();
	errx(2, "output may be truncated");
}
back to top