contrib/halog/halog.c

/*
 * haproxy log statistics reporter
 *
 * Copyright 2000-2010 Willy Tarreau <w@1wt.eu>
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License
 * as published by the Free Software Foundation; either version
 * 2 of the License, or (at your option) any later version.
 *
 */

#include <errno.h>
#include <fcntl.h>
#include <stdio.h>
#include <stdlib.h>
#include <syslog.h>
#include <string.h>
#include <unistd.h>
#include <ctype.h>

#include <eb32tree.h>
#include <eb64tree.h>
#include <ebistree.h>
#include <ebsttree.h>

#define SOURCE_FIELD 5
#define ACCEPT_FIELD 6
#define SERVER_FIELD 8
#define TIME_FIELD 9
#define STATUS_FIELD 10
#define BYTES_SENT_FIELD 11
#define TERM_CODES_FIELD 14
#define CONN_FIELD 15
#define QUEUE_LEN_FIELD 16
#define METH_FIELD 17
#define URL_FIELD 18
#define MAXLINE 16384
#define QBITS 4

#define SEP(c) ((unsigned char)(c) <= ' ')
#define SKIP_CHAR(p,c) do { while (1) { int __c = (unsigned char)*p++; if (__c == c) break; if (__c <= ' ') { p--; break; } } } while (0)

/* [0] = err/date, [1] = req, [2] = conn, [3] = resp, [4] = data */
static struct eb_root timers[5] = {
	EB_ROOT_UNIQUE, EB_ROOT_UNIQUE, EB_ROOT_UNIQUE,
	EB_ROOT_UNIQUE, EB_ROOT_UNIQUE,
};

struct timer {
	struct eb32_node node;
	unsigned int count;
};

struct srv_st {
	unsigned int st_cnt[6]; /* 0xx to 5xx */
	unsigned int nb_ct, nb_rt, nb_ok;
	unsigned long long cum_ct, cum_rt;
	struct ebmb_node node;
	/* don't put anything else here, the server name will be there */
};

struct url_stat {
	union {
		struct ebpt_node url;
		struct eb64_node val;
	} node;
	char *url;
	unsigned long long total_time;    /* sum(all reqs' times) */
	unsigned long long total_time_ok; /* sum(all OK reqs' times) */
	unsigned long long total_bytes_sent; /* sum(all bytes sent) */
	unsigned int nb_err, nb_req;
};

#define FILT_COUNT_ONLY		0x01
#define FILT_INVERT		0x02
#define FILT_QUIET		0x04
#define FILT_ERRORS_ONLY	0x08
#define FILT_ACC_DELAY		0x10
#define FILT_ACC_COUNT		0x20
#define FILT_GRAPH_TIMERS	0x40
#define FILT_PERCENTILE		0x80
#define FILT_TIME_RESP         0x100

#define FILT_INVERT_ERRORS     0x200
#define FILT_INVERT_TIME_RESP  0x400

#define FILT_COUNT_STATUS      0x800
#define FILT_COUNT_SRV_STATUS 0x1000
#define FILT_COUNT_TERM_CODES 0x2000

#define FILT_COUNT_URL_ONLY  0x004000
#define FILT_COUNT_URL_COUNT 0x008000
#define FILT_COUNT_URL_ERR   0x010000
#define FILT_COUNT_URL_TTOT  0x020000
#define FILT_COUNT_URL_TAVG  0x040000
#define FILT_COUNT_URL_TTOTO 0x080000
#define FILT_COUNT_URL_TAVGO 0x100000

#define FILT_HTTP_ONLY       0x200000
#define FILT_TERM_CODE_NAME  0x400000
#define FILT_INVERT_TERM_CODE_NAME 0x800000

#define FILT_HTTP_STATUS           0x1000000
#define FILT_INVERT_HTTP_STATUS    0x2000000
#define FILT_QUEUE_ONLY            0x4000000
#define FILT_QUEUE_SRV_ONLY        0x8000000

#define FILT_COUNT_URL_BAVG       0x10000000
#define FILT_COUNT_URL_BTOT       0x20000000

#define FILT_COUNT_URL_ANY   (FILT_COUNT_URL_ONLY|FILT_COUNT_URL_COUNT|FILT_COUNT_URL_ERR| \
			      FILT_COUNT_URL_TTOT|FILT_COUNT_URL_TAVG|FILT_COUNT_URL_TTOTO|FILT_COUNT_URL_TAVGO| \
			      FILT_COUNT_URL_BAVG|FILT_COUNT_URL_BTOT)

unsigned int filter = 0;
unsigned int filter_invert = 0;
const char *line;
int linenum = 0;
int parse_err = 0;
int lines_out = 0;

const char *fgets2(FILE *stream);

void filter_count_url(const char *accept_field, const char *time_field, struct timer **tptr);
void filter_count_srv_status(const char *accept_field, const char *time_field, struct timer **tptr);
void filter_count_term_codes(const char *accept_field, const char *time_field, struct timer **tptr);
void filter_count_status(const char *accept_field, const char *time_field, struct timer **tptr);
void filter_graphs(const char *accept_field, const char *time_field, struct timer **tptr);
void filter_output_line(const char *accept_field, const char *time_field, struct timer **tptr);
void filter_accept_holes(const char *accept_field, const char *time_field, struct timer **tptr);

void usage(FILE *output, const char *msg)
{
	fprintf(output,
		"%s"
		"Usage: halog [-h|--help] for long help\n"
		"       halog [-q] [-c]\n"
		"       {-gt|-pct|-st|-tc|-srv|-u|-uc|-ue|-ua|-ut|-uao|-uto|-uba|-ubt}\n"
		"       [-s <skip>] [-e|-E] [-H] [-rt|-RT <time>] [-ad <delay>] [-ac <count>]\n"
		"       [-v] [-Q|-QS] [-tcn|-TCN <termcode>] [ -hs|-HS [min][:[max]] ] < log\n"
		"\n",
		msg ? msg : ""
		);
}

void die(const char *msg)
{
	usage(stderr, msg);
	exit(1);
}

void help()
{
	usage(stdout, NULL);
	printf(
	       "Input filters (several filters may be combined) :\n"
	       " -H                      only match lines containing HTTP logs (ignore TCP)\n"
	       " -E                      only match lines without any error (no 5xx status)\n"
	       " -e                      only match lines with errors (status 5xx or negative)\n"
	       " -rt|-RT <time>          only match response times larger|smaller than <time>\n"
	       " -Q|-QS                  only match queued requests (any queue|server queue)\n"
	       " -tcn|-TCN <code>        only match requests with/without termination code <code>\n"
	       " -hs|-HS <[min][:][max]> only match requests with HTTP status codes within/not\n"
	       "                         within min..max. Any of them may be omitted. Exact\n"
	       "                         code is checked for if no ':' is specified.\n"
	       "Modifiers\n"
	       " -v                      invert the input filtering condition\n"
	       " -q                      don't report errors/warnings\n"
	       "\n"
	       "Output filters - only one may be used at a time\n"
	       " -c    only report the number of lines that would have been printed\n"
	       " -pct  output connect and response times percentiles\n"
	       " -st   output number of requests per HTTP status code\n"
	       " -tc   output number of requests per termination code (2 chars)\n"
	       " -srv  output statistics per server (time, requests, errors)\n"
	       " -u*   output statistics per URL (time, requests, errors)\n"
	       "       Additional characters indicate the output sorting key :\n"
	       "       -u : by URL, -uc : request count, -ue : error count\n"
	       "       -ua : average response time, -uto : average total time\n"
	       "       -uao, -uto: average times computed on valid ('OK') requests\n"
	       "       -uba, -ubt: average bytes returned, total bytes returned\n"
	       );
	exit(0);
}


/* return pointer to first char not part of current field starting at <p>. */

#if defined(__i386__)
/* this one is always faster on 32-bits */
static inline const char *field_stop(const char *p)
{
	asm(
	    /* Look for spaces */
	    "4:                  \n\t"
	    "inc   %0            \n\t"
	    "cmpb  $0x20, -1(%0) \n\t"
	    "ja    4b            \n\t"
	    "jz    3f            \n\t"

	    /* we only get there for control chars 0..31. Leave if we find '\0' */
	    "cmpb  $0x0, -1(%0)  \n\t"
	    "jnz   4b            \n\t"

	    /* return %0-1 = position of the last char we checked */
	    "3:                  \n\t"
	    "dec   %0            \n\t"
	    : "=r" (p)
	    : "0" (p)
	    );
	return p;
}
#else
const char *field_stop(const char *p)
{
	unsigned char c;

	while (1) {
		c = *(p++);
		if (c > ' ')
			continue;
		if (c == ' ' || c == 0)
			break;
	}
	return p - 1;
}
#endif

/* return field <field> (starting from 1) in string <p>. Only consider
 * contiguous spaces (or tabs) as one delimiter. May return pointer to
 * last char if field is not found. Equivalent to awk '{print $field}'.
 */
const char *field_start(const char *p, int field)
{
#ifndef PREFER_ASM
	unsigned char c;
	while (1) {
		/* skip spaces */
		while (1) {
			c = *(p++);
			if (c > ' ')
				break;
			if (c == ' ')
				continue;
			if (!c) /* end of line */
				return p-1;
			/* other char => new field */
			break;
		}

		/* start of field */
		field--;
		if (!field)
			return p-1;

		/* skip this field */
		while (1) {
			c = *(p++);
			if (c == ' ')
				break;
			if (c > ' ')
				continue;
			if (c == '\0')
				return p - 1;
		}
	}
#else
	/* This version works optimally on i386 and x86_64 but the code above
	 * shows similar performance. However, depending on the version of GCC
	 * used, inlining rules change and it may have difficulties to make
	 * efficient use of this code at other locations and could result in
	 * worse performance (eg: gcc 4.4). You may want to experience.
	 */
	asm(
	    /* skip spaces */
	    "1:                  \n\t"
	    "inc   %0            \n\t"
	    "cmpb  $0x20, -1(%0) \n\t"
	    "ja    2f            \n\t"
	    "jz    1b            \n\t"

	    /* we only get there for control chars 0..31. Leave if we find '\0' */
	    "cmpb  $0x0, -1(%0)  \n\t"
	    "jz    3f            \n\t"

	    /* start of field at [%0-1]. Check if we need to skip more fields */
	    "2:                  \n\t"
	    "dec   %1            \n\t"
	    "jz    3f            \n\t"

	    /* Look for spaces */
	    "4:                  \n\t"
	    "inc   %0            \n\t"
	    "cmpb  $0x20, -1(%0) \n\t"
	    "jz    1b            \n\t"
	    "ja    4b            \n\t"

	    /* we only get there for control chars 0..31. Leave if we find '\0' */
	    "cmpb  $0x0, -1(%0)  \n\t"
	    "jnz   4b            \n\t"

	    /* return %0-1 = position of the last char we checked */
	    "3:                  \n\t"
	    "dec   %0            \n\t"
	    : "=r" (p)
	    : "r" (field), "0" (p)
	    );
	return p;
#endif
}

/* keep only the <bits> higher bits of <i> */
static inline unsigned int quantify_u32(unsigned int i, int bits)
{
	int high;

	if (!bits)
		return 0;

	if (i)
		high = fls_auto(i);    // 1 to 32
	else
		high = 0;

	if (high <= bits)
		return i;

	return i & ~((1 << (high - bits)) - 1);
}

/* keep only the <bits> higher bits of the absolute value of <i>, as well as
 * its sign. */
static inline int quantify(int i, int bits)
{
	if (i >= 0)
		return quantify_u32(i, bits);
	else
		return -quantify_u32(-i, bits);
}

/* Insert timer value <v> into tree <r>. A pre-allocated node must be passed
 * in <alloc>. It may be NULL, in which case the function will allocate it
 * itself. It will be reset to NULL once consumed. The caller is responsible
 * for freeing the node once not used anymore. The node where the value was
 * inserted is returned.
 */
struct timer *insert_timer(struct eb_root *r, struct timer **alloc, int v)
{
	struct timer *t = *alloc;
	struct eb32_node *n;

	if (!t) {
		t = calloc(sizeof(*t), 1);
		if (unlikely(!t)) {
			fprintf(stderr, "%s: not enough memory\n", __FUNCTION__);
			exit(1);
		}
	}
	t->node.key = quantify(v, QBITS);         // keep only the higher QBITS bits

	n = eb32i_insert(r, &t->node);
	if (n == &t->node)
		t = NULL;  /* node inserted, will malloc next time */

	*alloc = t;
	return container_of(n, struct timer, node);
}

/* Insert value value <v> into tree <r>. A pre-allocated node must be passed
 * in <alloc>. It may be NULL, in which case the function will allocate it
 * itself. It will be reset to NULL once consumed. The caller is responsible
 * for freeing the node once not used anymore. The node where the value was
 * inserted is returned.
 */
struct timer *insert_value(struct eb_root *r, struct timer **alloc, int v)
{
	struct timer *t = *alloc;
	struct eb32_node *n;

	if (!t) {
		t = calloc(sizeof(*t), 1);
		if (unlikely(!t)) {
			fprintf(stderr, "%s: not enough memory\n", __FUNCTION__);
			exit(1);
		}
	}
	t->node.key = v;

	n = eb32i_insert(r, &t->node);
	if (n == &t->node)
		t = NULL;  /* node inserted, will malloc next time */

	*alloc = t;
	return container_of(n, struct timer, node);
}

int str2ic(const char *s)
{
	int i = 0;
	int j, k;

	if (*s != '-') {
		/* positive number */
		while (1) {
			j = (*s++) - '0';
			k = i * 10;
			if ((unsigned)j > 9)
				break;
			i = k + j;
		}
	} else {
		/* negative number */
		s++;
		while (1) {
			j = (*s++) - '0';
			k = i * 10;
			if ((unsigned)j > 9)
				break;
			i = k - j;
		}
	}

	return i;
}


/* Equivalent to strtoul with a length. */
static inline unsigned int __strl2ui(const char *s, int len)
{
	unsigned int i = 0;
	while (len-- > 0) {
		i = i * 10 - '0';
		i += (unsigned char)*s++;
	}
	return i;
}

unsigned int strl2ui(const char *s, int len)
{
	return __strl2ui(s, len);
}

/* Convert "[04/Dec/2008:09:49:40.555]" to an integer equivalent to the time of
 * the day in milliseconds. It returns -1 for all unparsable values. The parser
 * looks ugly but gcc emits far better code that way.
 */
int convert_date(const char *field)
{
	unsigned int h, m, s, ms;
	unsigned char c;
	const char *b, *e;

	h = m = s = ms = 0;
	e = field;

	/* skip the date */
	while (1) {
		c = *(e++);
		if (c == ':')
			break;
		if (!c)
			goto out_err;
	}

	/* hour + ':' */
	b = e;
	while (1) {
		c = *(e++) - '0';
		if (c > 9)
			break;
		h = h * 10 + c;
	}
	if (c == (unsigned char)(0 - '0'))
		goto out_err;

	/* minute + ':' */
	b = e;
	while (1) {
		c = *(e++) - '0';
		if (c > 9)
			break;
		m = m * 10 + c;
	}
	if (c == (unsigned char)(0 - '0'))
		goto out_err;

	/* second + '.' or ']' */
	b = e;
	while (1) {
		c = *(e++) - '0';
		if (c > 9)
			break;
		s = s * 10 + c;
	}
	if (c == (unsigned char)(0 - '0'))
		goto out_err;

	/* if there's a '.', we have milliseconds */
	if (c == (unsigned char)('.' - '0')) {
		/* millisecond second + ']' */
		b = e;
		while (1) {
			c = *(e++) - '0';
			if (c > 9)
				break;
			ms = ms * 10 + c;
		}
		if (c == (unsigned char)(0 - '0'))
			goto out_err;
	}
	return (((h * 60) + m) * 60 + s) * 1000 + ms;
 out_err:
	return -1;
}

void truncated_line(int linenum, const char *line)
{
	if (!(filter & FILT_QUIET))
		fprintf(stderr, "Truncated line %d: %s\n", linenum, line);
}

int main(int argc, char **argv)
{
	const char *b, *e, *p, *time_field, *accept_field;
	const char *filter_term_code_name = NULL;
	const char *output_file = NULL;
	int f, last, err;
	struct timer *t = NULL;
	struct eb32_node *n;
	struct url_stat *ustat = NULL;
	int val, test;
	int filter_acc_delay = 0, filter_acc_count = 0;
	int filter_time_resp = 0;
	int filt_http_status_low = 0, filt_http_status_high = 0;
	int skip_fields = 1;

	void (*line_filter)(const char *accept_field, const char *time_field, struct timer **tptr) = NULL;

	argc--; argv++;
	while (argc > 0) {
		if (*argv[0] != '-')
			break;

		if (strcmp(argv[0], "-ad") == 0) {
			if (argc < 2) die("missing option for -ad");
			argc--; argv++;
			filter |= FILT_ACC_DELAY;
			filter_acc_delay = atol(*argv);
		}
		else if (strcmp(argv[0], "-ac") == 0) {
			if (argc < 2) die("missing option for -ac");
			argc--; argv++;
			filter |= FILT_ACC_COUNT;
			filter_acc_count = atol(*argv);
		}
		else if (strcmp(argv[0], "-rt") == 0) {
			if (argc < 2) die("missing option for -rt");
			argc--; argv++;
			filter |= FILT_TIME_RESP;
			filter_time_resp = atol(*argv);
		}
		else if (strcmp(argv[0], "-RT") == 0) {
			if (argc < 2) die("missing option for -RT");
			argc--; argv++;
			filter |= FILT_TIME_RESP | FILT_INVERT_TIME_RESP;
			filter_time_resp = atol(*argv);
		}
		else if (strcmp(argv[0], "-s") == 0) {
			if (argc < 2) die("missing option for -s");
			argc--; argv++;
			skip_fields = atol(*argv);
		}
		else if (strcmp(argv[0], "-e") == 0)
			filter |= FILT_ERRORS_ONLY;
		else if (strcmp(argv[0], "-E") == 0)
			filter |= FILT_ERRORS_ONLY | FILT_INVERT_ERRORS;
		else if (strcmp(argv[0], "-H") == 0)
			filter |= FILT_HTTP_ONLY;
		else if (strcmp(argv[0], "-Q") == 0)
			filter |= FILT_QUEUE_ONLY;
		else if (strcmp(argv[0], "-QS") == 0)
			filter |= FILT_QUEUE_SRV_ONLY;
		else if (strcmp(argv[0], "-c") == 0)
			filter |= FILT_COUNT_ONLY;
		else if (strcmp(argv[0], "-q") == 0)
			filter |= FILT_QUIET;
		else if (strcmp(argv[0], "-v") == 0)
			filter_invert = !filter_invert;
		else if (strcmp(argv[0], "-gt") == 0)
			filter |= FILT_GRAPH_TIMERS;
		else if (strcmp(argv[0], "-pct") == 0)
			filter |= FILT_PERCENTILE;
		else if (strcmp(argv[0], "-st") == 0)
			filter |= FILT_COUNT_STATUS;
		else if (strcmp(argv[0], "-srv") == 0)
			filter |= FILT_COUNT_SRV_STATUS;
		else if (strcmp(argv[0], "-tc") == 0)
			filter |= FILT_COUNT_TERM_CODES;
		else if (strcmp(argv[0], "-tcn") == 0) {
			if (argc < 2) die("missing option for -tcn");
			argc--; argv++;
			filter |= FILT_TERM_CODE_NAME;
			filter_term_code_name = *argv;
		}
		else if (strcmp(argv[0], "-TCN") == 0) {
			if (argc < 2) die("missing option for -TCN");
			argc--; argv++;
			filter |= FILT_TERM_CODE_NAME | FILT_INVERT_TERM_CODE_NAME;
			filter_term_code_name = *argv;
		}
		else if (strcmp(argv[0], "-hs") == 0 || strcmp(argv[0], "-HS") == 0) {
			char *sep, *str;

			if (argc < 2) die("missing option for -hs/-HS ([min]:[max])");
			filter |= FILT_HTTP_STATUS;
			if (argv[0][1] == 'H')
				filter |= FILT_INVERT_HTTP_STATUS;

			argc--; argv++;
			str = *argv;
			sep = strchr(str, ':');  /* [min]:[max] */
			if (!sep)
				sep = str; /* make max point to min */
			else
				*sep++ = 0;
			filt_http_status_low = *str ? atol(str) : 0;
			filt_http_status_high = *sep ? atol(sep) : 65535;
		}
		else if (strcmp(argv[0], "-u") == 0)
			filter |= FILT_COUNT_URL_ONLY;
		else if (strcmp(argv[0], "-uc") == 0)
			filter |= FILT_COUNT_URL_COUNT;
		else if (strcmp(argv[0], "-ue") == 0)
			filter |= FILT_COUNT_URL_ERR;
		else if (strcmp(argv[0], "-ua") == 0)
			filter |= FILT_COUNT_URL_TAVG;
		else if (strcmp(argv[0], "-ut") == 0)
			filter |= FILT_COUNT_URL_TTOT;
		else if (strcmp(argv[0], "-uao") == 0)
			filter |= FILT_COUNT_URL_TAVGO;
		else if (strcmp(argv[0], "-uto") == 0)
			filter |= FILT_COUNT_URL_TTOTO;
		else if (strcmp(argv[0], "-uba") == 0)
			filter |= FILT_COUNT_URL_BAVG;
		else if (strcmp(argv[0], "-ubt") == 0)
			filter |= FILT_COUNT_URL_BTOT;
		else if (strcmp(argv[0], "-o") == 0) {
			if (output_file)
				die("Fatal: output file name already specified.\n");
			if (argc < 2)
				die("Fatal: missing output file name.\n");
			output_file = argv[1];
		}
		else if (strcmp(argv[0], "-h") == 0 || strcmp(argv[0], "--help") == 0)
			help();
		argc--;
		argv++;
	}

	if (!filter)
		die("No action specified.\n");

	if (filter & FILT_ACC_COUNT && !filter_acc_count)
		filter_acc_count=1;

	if (filter & FILT_ACC_DELAY && !filter_acc_delay)
		filter_acc_delay = 1;


	/* by default, all lines are printed */
	line_filter = filter_output_line;
	if (filter & (FILT_ACC_COUNT|FILT_ACC_DELAY))
		line_filter = filter_accept_holes;
	else if (filter & (FILT_GRAPH_TIMERS|FILT_PERCENTILE))
		line_filter = filter_graphs;
	else if (filter & FILT_COUNT_STATUS)
		line_filter = filter_count_status;
	else if (filter & FILT_COUNT_TERM_CODES)
		line_filter = filter_count_term_codes;
	else if (filter & FILT_COUNT_SRV_STATUS)
		line_filter = filter_count_srv_status;
	else if (filter & FILT_COUNT_URL_ANY)
		line_filter = filter_count_url;
	else if (filter & FILT_COUNT_ONLY)
		line_filter = NULL;

#if defined(POSIX_FADV_SEQUENTIAL)
	/* around 20% performance improvement is observed on Linux with this
	 * on cold-cache. Surprizingly, WILLNEED is less performant. Don't
	 * use NOREUSE as it flushes the cache and prevents easy data
	 * manipulation on logs!
	 */
	posix_fadvise(0, 0, 0, POSIX_FADV_SEQUENTIAL);
#endif

	if (!line_filter &&
	    !(filter & (FILT_HTTP_ONLY|FILT_TIME_RESP|FILT_ERRORS_ONLY|FILT_HTTP_STATUS|FILT_QUEUE_ONLY|FILT_QUEUE_SRV_ONLY|FILT_TERM_CODE_NAME))) {
		/* read the whole file at once first */
		if (!filter_invert)
			while (fgets2(stdin) != NULL)
				lines_out++;

		goto skip_filters;
	}

	while ((line = fgets2(stdin)) != NULL) {
		linenum++;
		time_field = NULL; accept_field = NULL;

		test = 1;

		/* for any line we process, we first ensure that there is a field
		 * looking like the accept date field (beginning with a '[').
		 */
		accept_field = field_start(line, ACCEPT_FIELD + skip_fields);
		if (unlikely(*accept_field != '[')) {
			parse_err++;
			continue;
		}

		/* the day of month field is begin 01 and 31 */
		if (accept_field[1] < '0' || accept_field[1] > '3') {
			parse_err++;
			continue;
		}

		if (filter & FILT_HTTP_ONLY) {
			/* only report lines with at least 4 timers */
			if (!time_field) {
				time_field = field_start(accept_field, TIME_FIELD - ACCEPT_FIELD + 1);
				if (unlikely(!*time_field)) {
					truncated_line(linenum, line);
					continue;
				}
			}

			e = field_stop(time_field + 1);
			/* we have field TIME_FIELD in [time_field]..[e-1] */
			p = time_field;
			f = 0;
			while (!SEP(*p)) {
				if (++f == 4)
					break;
				SKIP_CHAR(p, '/');
			}
			test &= (f >= 4);
		}

		if (filter & FILT_TIME_RESP) {
			int tps;

			/* only report lines with response times larger than filter_time_resp */
			if (!time_field) {
				time_field = field_start(accept_field, TIME_FIELD - ACCEPT_FIELD + 1);
				if (unlikely(!*time_field)) {
					truncated_line(linenum, line);
					continue;
				}
			}

			e = field_stop(time_field + 1);
			/* we have field TIME_FIELD in [time_field]..[e-1], let's check only the response time */

			p = time_field;
			err = 0;
			f = 0;
			while (!SEP(*p)) {
				tps = str2ic(p);
				if (tps < 0) {
					tps = -1;
					err = 1;
				}
				if (++f == 4)
					break;
				SKIP_CHAR(p, '/');
			}

			if (unlikely(f < 4)) {
				parse_err++;
				continue;
			}

			test &= (tps >= filter_time_resp) ^ !!(filter & FILT_INVERT_TIME_RESP);
		}

		if (filter & (FILT_ERRORS_ONLY | FILT_HTTP_STATUS)) {
			/* Check both error codes (-1, 5xx) and status code ranges */
			if (time_field)
				b = field_start(time_field, STATUS_FIELD - TIME_FIELD + 1);
			else
				b = field_start(accept_field, STATUS_FIELD - ACCEPT_FIELD + 1);

			if (unlikely(!*b)) {
				truncated_line(linenum, line);
				continue;
			}

			val = str2ic(b);
			if (filter & FILT_ERRORS_ONLY)
				test &= (val < 0 || (val >= 500 && val <= 599)) ^ !!(filter & FILT_INVERT_ERRORS);

			if (filter & FILT_HTTP_STATUS)
				test &= (val >= filt_http_status_low && val <= filt_http_status_high) ^ !!(filter & FILT_INVERT_HTTP_STATUS);
		}

		if (filter & (FILT_QUEUE_ONLY|FILT_QUEUE_SRV_ONLY)) {
			/* Check if the server's queue is non-nul */
			if (time_field)
				b = field_start(time_field, QUEUE_LEN_FIELD - TIME_FIELD + 1);
			else
				b = field_start(accept_field, QUEUE_LEN_FIELD - ACCEPT_FIELD + 1);

			if (unlikely(!*b)) {
				truncated_line(linenum, line);
				continue;
			}

			if (*b == '0') {
				if (filter & FILT_QUEUE_SRV_ONLY) {
					test = 0;
				}
				else {
					do {
						b++;
						if (*b == '/') {
							b++;
							break;
						}
					} while (*b);
					test &= ((unsigned char)(*b - '1') < 9);
				}
			}
		}

		if (filter & FILT_TERM_CODE_NAME) {
			/* only report corresponding termination code name */
			if (time_field)
				b = field_start(time_field, TERM_CODES_FIELD - TIME_FIELD + 1);
			else
				b = field_start(accept_field, TERM_CODES_FIELD - ACCEPT_FIELD + 1);

			if (unlikely(!*b)) {
				truncated_line(linenum, line);
				continue;
			}

			test &= (b[0] == filter_term_code_name[0] && b[1] == filter_term_code_name[1]) ^ !!(filter & FILT_INVERT_TERM_CODE_NAME);
		}


		test ^= filter_invert;
		if (!test)
			continue;

		/************** here we process inputs *******************/

		if (line_filter)
			line_filter(accept_field, time_field, &t);
		else
			lines_out++; /* we're just counting lines */
	}

 skip_filters:
	/*****************************************************
	 * Here we've finished reading all input. Depending on the
	 * filters, we may still have some analysis to run on the
	 * collected data and to output data in a new format.
	 *************************************************** */

	if (t)
		free(t);

	if (filter & FILT_COUNT_ONLY) {
		printf("%d\n", lines_out);
		exit(0);
	}

	if (filter & (FILT_ACC_COUNT|FILT_ACC_DELAY)) {
		/* sort and count all timers. Output will look like this :
		 * <accept_date> <delta_ms from previous one> <nb entries>
		 */
		n = eb32_first(&timers[0]);

		if (n)
			last = n->key;
		while (n) {
			unsigned int d, h, m, s, ms;

			t = container_of(n, struct timer, node);
			h = n->key;
			d = h - last;
			last = h;

			if (d >= filter_acc_delay && t->count >= filter_acc_count) {
				ms = h % 1000; h = h / 1000;
				s = h % 60; h = h / 60;
				m = h % 60; h = h / 60;
				lines_out++;
				printf("%02d:%02d:%02d.%03d %d %d %d\n", h, m, s, ms, last, d, t->count);
			}
			n = eb32_next(n);
		}
	}
	else if (filter & FILT_GRAPH_TIMERS) {
		/* sort all timers */
		for (f = 0; f < 5; f++) {
			struct eb32_node *n;
			int val;

			val = 0;
			n = eb32_first(&timers[f]);
			while (n) {
				int i;
				double d;

				t = container_of(n, struct timer, node);
				last = n->key;
				val = t->count;

				i = (last < 0) ? -last : last;
				i = fls_auto(i) - QBITS;

				if (i > 0)
					d = val / (double)(1 << i);
				else
					d = val;

				if (d > 0.0) {
					printf("%d %d %f\n", f, last, d+1.0);
					lines_out++;
				}

				n = eb32_next(n);
			}
		}
	}
	else if (filter & FILT_PERCENTILE) {
		/* report timers by percentile :
		 *    <percent> <total> <max_req_time> <max_conn_time> <max_resp_time> <max_data_time>
		 * We don't count errs.
		 */
		struct eb32_node *n[5];
		unsigned long cum[5];
		double step;

		if (!lines_out)
			goto empty;

		for (f = 1; f < 5; f++) {
			n[f] = eb32_first(&timers[f]);
			cum[f] = container_of(n[f], struct timer, node)->count;
		}

		for (step = 1; step <= 1000;) {
			unsigned int thres = lines_out * (step / 1000.0);

			printf("%3.1f %d ", step/10.0, thres);
			for (f = 1; f < 5; f++) {
				struct eb32_node *next;
				while (cum[f] < thres) {
					/* need to find other keys */
					next = eb32_next(n[f]);
					if (!next)
						break;
					n[f] = next;
					cum[f] += container_of(next, struct timer, node)->count;
				}

				/* value still within $step % of total */
				printf("%d ", n[f]->key);
			}
			putchar('\n');
			if (step >= 100 && step < 900)
				step += 50;  // jump 5% by 5% between those steps.
			else if (step >= 20 && step < 980)
				step += 10;
			else
				step += 1;
		}
	}
	else if (filter & FILT_COUNT_STATUS) {
		/* output all statuses in the form of <status> <occurrences> */
		n = eb32_first(&timers[0]);
		while (n) {
			t = container_of(n, struct timer, node);
			printf("%d %d\n", n->key, t->count);
			lines_out++;
			n = eb32_next(n);
		}
	}
	else if (filter & FILT_COUNT_SRV_STATUS) {
		struct ebmb_node *srv_node;
		struct srv_st *srv;

		printf("#srv_name 1xx 2xx 3xx 4xx 5xx other tot_req req_ok pct_ok avg_ct avg_rt\n");

		srv_node = ebmb_first(&timers[0]);
		while (srv_node) {
			int tot_rq;

			srv = container_of(srv_node, struct srv_st, node);

			tot_rq = 0;
			for (f = 0; f <= 5; f++)
				tot_rq += srv->st_cnt[f];

			printf("%s %d %d %d %d %d %d %d %d %.1f %d %d\n",
			       srv_node->key, srv->st_cnt[1], srv->st_cnt[2],
			       srv->st_cnt[3], srv->st_cnt[4], srv->st_cnt[5], srv->st_cnt[0],
			       tot_rq,
			       srv->nb_ok, (double)srv->nb_ok * 100.0 / (tot_rq?tot_rq:1),
			       (int)(srv->cum_ct / (srv->nb_ct?srv->nb_ct:1)), (int)(srv->cum_rt / (srv->nb_rt?srv->nb_rt:1)));
			srv_node = ebmb_next(srv_node);
			lines_out++;
		}
	}
	else if (filter & FILT_COUNT_TERM_CODES) {
		/* output all statuses in the form of <code> <occurrences> */
		n = eb32_first(&timers[0]);
		while (n) {
			t = container_of(n, struct timer, node);
			printf("%c%c %d\n", (n->key >> 8), (n->key) & 255, t->count);
			lines_out++;
			n = eb32_next(n);
		}
	}
	else if (filter & FILT_COUNT_URL_ANY) {
		struct eb_node *node, *next;

		if (!(filter & FILT_COUNT_URL_ONLY)) {
			/* we have to sort on another criterion. We'll use timers[1] for the
			 * destination tree.
			 */

			timers[1] = EB_ROOT; /* reconfigure to accept duplicates */
			for (node = eb_first(&timers[0]); node; node = next) {
				next = eb_next(node);
				eb_delete(node);

				ustat = container_of(node, struct url_stat, node.url.node);

				if (filter & FILT_COUNT_URL_COUNT)
					ustat->node.val.key = ustat->nb_req;
				else if (filter & FILT_COUNT_URL_ERR)
					ustat->node.val.key = ustat->nb_err;
				else if (filter & FILT_COUNT_URL_TTOT)
					ustat->node.val.key = ustat->total_time;
				else if (filter & FILT_COUNT_URL_TAVG)
					ustat->node.val.key = ustat->nb_req ? ustat->total_time / ustat->nb_req : 0;
				else if (filter & FILT_COUNT_URL_TTOTO)
					ustat->node.val.key = ustat->total_time_ok;
				else if (filter & FILT_COUNT_URL_TAVGO)
					ustat->node.val.key = (ustat->nb_req - ustat->nb_err) ? ustat->total_time_ok / (ustat->nb_req - ustat->nb_err) : 0;
				else if (filter & FILT_COUNT_URL_BAVG)
					ustat->node.val.key = ustat->nb_req ? ustat->total_bytes_sent / ustat->nb_req : 0;
				else if (filter & FILT_COUNT_URL_BTOT)
					ustat->node.val.key = ustat->total_bytes_sent;
				else
					ustat->node.val.key = 0;

				eb64_insert(&timers[1], &ustat->node.val);
			}
			/* switch trees */
			timers[0] = timers[1];
		}

		printf("#req err ttot tavg oktot okavg bavg btot url\n");

		/* scan the tree in its reverse sorting order */
		node = eb_last(&timers[0]);
		while (node) {
			ustat = container_of(node, struct url_stat, node.url.node);
			printf("%d %d %Ld %Ld %Ld %Ld %Ld %Ld %s\n",
			       ustat->nb_req,
			       ustat->nb_err,
			       ustat->total_time,
			       ustat->nb_req ? ustat->total_time / ustat->nb_req : 0,
			       ustat->total_time_ok,
			       (ustat->nb_req - ustat->nb_err) ? ustat->total_time_ok / (ustat->nb_req - ustat->nb_err) : 0,
			       ustat->nb_req ? ustat->total_bytes_sent / ustat->nb_req : 0,
			       ustat->total_bytes_sent,
			       ustat->url);

			node = eb_prev(node);
			lines_out++;
		}
	}

 empty:
	if (!(filter & FILT_QUIET))
		fprintf(stderr, "%d lines in, %d lines out, %d parsing errors\n",
			linenum, lines_out, parse_err);
	exit(0);
}

void filter_output_line(const char *accept_field, const char *time_field, struct timer **tptr)
{
	puts(line);
	lines_out++;
}

void filter_accept_holes(const char *accept_field, const char *time_field, struct timer **tptr)
{
	struct timer *t2;
	int val;

	val = convert_date(accept_field);
	if (unlikely(val < 0)) {
		truncated_line(linenum, line);
		return;
	}

	t2 = insert_value(&timers[0], tptr, val);
	t2->count++;
	lines_out++;
	return;
}

void filter_count_status(const char *accept_field, const char *time_field, struct timer **tptr)
{
	struct timer *t2;
	const char *b;
	int val;

	if (time_field)
		b = field_start(time_field, STATUS_FIELD - TIME_FIELD + 1);
	else
		b = field_start(accept_field, STATUS_FIELD - ACCEPT_FIELD + 1);

	if (unlikely(!*b)) {
		truncated_line(linenum, line);
		return;
	}

	val = str2ic(b);

	t2 = insert_value(&timers[0], tptr, val);
	t2->count++;
}

void filter_count_term_codes(const char *accept_field, const char *time_field, struct timer **tptr)
{
	struct timer *t2;
	const char *b;
	int val;

	if (time_field)
		b = field_start(time_field, TERM_CODES_FIELD - TIME_FIELD + 1);
	else
		b = field_start(accept_field, TERM_CODES_FIELD - ACCEPT_FIELD + 1);

	if (unlikely(!*b)) {
		truncated_line(linenum, line);
		return;
	}

	val = 256 * b[0] + b[1];

	t2 = insert_value(&timers[0], tptr, val);
	t2->count++;
}

void filter_count_srv_status(const char *accept_field, const char *time_field, struct timer **tptr)
{
	const char *b, *e, *p;
	int f, err, array[5];
	struct ebmb_node *srv_node;
	struct srv_st *srv;
	int val;

	/* the server field is before the status field, so let's
	 * parse them in the proper order.
	 */
	b = field_start(accept_field, SERVER_FIELD - ACCEPT_FIELD + 1);
	if (unlikely(!*b)) {
		truncated_line(linenum, line);
		return;
	}

	e = field_stop(b + 1);  /* we have the server name in [b]..[e-1] */

	/* the chance that a server name already exists is extremely high,
	 * so let's perform a normal lookup first.
	 */
	srv_node = ebst_lookup_len(&timers[0], b, e - b);
	srv = container_of(srv_node, struct srv_st, node);

	if (!srv_node) {
		/* server not yet in the tree, let's create it */
		srv = (void *)calloc(1, sizeof(struct srv_st) + e - b + 1);
		srv_node = &srv->node;
		memcpy(&srv_node->key, b, e - b);
		srv_node->key[e - b] = '\0';
		ebst_insert(&timers[0], srv_node);
	}

	/* let's collect the connect and response times */
	if (!time_field) {
		time_field = field_start(e, TIME_FIELD - SERVER_FIELD);
		if (unlikely(!*time_field)) {
			truncated_line(linenum, line);
			return;
		}
	}

	e = field_stop(time_field + 1);
	/* we have field TIME_FIELD in [time_field]..[e-1] */

	p = time_field;
	err = 0;
	f = 0;
	while (!SEP(*p)) {
		array[f] = str2ic(p);
		if (array[f] < 0) {
			array[f] = -1;
			err = 1;
		}
		if (++f == 5)
			break;
		SKIP_CHAR(p, '/');
	}

	if (unlikely(f < 5)){
		parse_err++;
		return;
	}

	/* OK we have our timers in array[2,3] */
	if (!err)
		srv->nb_ok++;

	if (array[2] >= 0) {
		srv->cum_ct += array[2];
		srv->nb_ct++;
	}

	if (array[3] >= 0) {
		srv->cum_rt += array[3];
		srv->nb_rt++;
	}

	/* we're interested in the 5 HTTP status classes (1xx ... 5xx), and
	 * the invalid ones which will be reported as 0.
	 */
	b = field_start(e, STATUS_FIELD - TIME_FIELD);
	if (unlikely(!*b)) {
		truncated_line(linenum, line);
		return;
	}

	val = 0;
	if (*b >= '1' && *b <= '5')
		val = *b - '0';

	srv->st_cnt[val]++;
}

void filter_count_url(const char *accept_field, const char *time_field, struct timer **tptr)
{
	struct url_stat *ustat = NULL;
	struct ebpt_node *ebpt_old;
	const char *b, *e;
	int f, err, array[5];
	int val;

	/* let's collect the response time */
	if (!time_field) {
		time_field = field_start(accept_field, TIME_FIELD - ACCEPT_FIELD + 1);  // avg 115 ns per line
		if (unlikely(!*time_field)) {
			truncated_line(linenum, line);
			return;
		}
	}

	/* we have the field TIME_FIELD starting at <time_field>. We'll
	 * parse the 5 timers to detect errors, it takes avg 55 ns per line.
	 */
	e = time_field; err = 0; f = 0;
	while (!SEP(*e)) {
		array[f] = str2ic(e);
		if (array[f] < 0) {
			array[f] = -1;
			err = 1;
		}
		if (++f == 5)
			break;
		SKIP_CHAR(e, '/');
	}
	if (f < 5) {
		parse_err++;
		return;
	}

	/* OK we have our timers in array[3], and err is >0 if at
	 * least one -1 was seen. <e> points to the first char of
	 * the last timer. Let's prepare a new node with that.
	 */
	if (unlikely(!ustat))
		ustat = calloc(1, sizeof(*ustat));

	ustat->nb_err = err;
	ustat->nb_req = 1;

	/* use array[4] = total time in case of error */
	ustat->total_time = (array[3] >= 0) ? array[3] : array[4];
	ustat->total_time_ok = (array[3] >= 0) ? array[3] : 0;

	e = field_start(e, BYTES_SENT_FIELD - TIME_FIELD + 1);
	val = str2ic(e);
	ustat->total_bytes_sent = val;

	/* the line may be truncated because of a bad request or anything like this,
	 * without a method. Also, if it does not begin with an quote, let's skip to
	 * the next field because it's a capture. Let's fall back to the "method" itself
	 * if there's nothing else.
	 */
	e = field_start(e, METH_FIELD - BYTES_SENT_FIELD + 1);
	while (*e != '"' && *e) {
		/* Note: some syslog servers escape quotes ! */
		if (*e == '\\' && e[1] == '"')
			break;
		e = field_start(e, 2);
	}

	if (unlikely(!*e)) {
		truncated_line(linenum, line);
		return;
	}

	b = field_start(e, URL_FIELD - METH_FIELD + 1); // avg 40 ns per line
	if (!*b)
		b = e;

	/* stop at end of field or first ';' or '?', takes avg 64 ns per line */
	e = b;
	do {
		if (*e == ' ' || *e == '?' || *e == ';') {
			*(char *)e = 0;
			break;
		}
		e++;
	} while (*e);

	/* now instead of copying the URL for a simple lookup, we'll link
	 * to it from the node we're trying to insert. If it returns a
	 * different value, it was already there. Otherwise we just have
	 * to dynamically realloc an entry using strdup().
	 */
	ustat->node.url.key = (char *)b;
	ebpt_old = ebis_insert(&timers[0], &ustat->node.url);

	if (ebpt_old != &ustat->node.url) {
		struct url_stat *ustat_old;
		/* node was already there, let's update previous one */
		ustat_old = container_of(ebpt_old, struct url_stat, node.url);
		ustat_old->nb_req ++;
		ustat_old->nb_err += ustat->nb_err;
		ustat_old->total_time += ustat->total_time;
		ustat_old->total_time_ok += ustat->total_time_ok;
		ustat_old->total_bytes_sent += ustat->total_bytes_sent;
	} else {
		ustat->url = ustat->node.url.key = strdup(ustat->node.url.key);
		ustat = NULL; /* node was used */
	}
}

void filter_graphs(const char *accept_field, const char *time_field, struct timer **tptr)
{
	struct timer *t2;
	const char *e, *p;
	int f, err, array[5];

	if (!time_field) {
		time_field = field_start(accept_field, TIME_FIELD - ACCEPT_FIELD + 1);
		if (unlikely(!*time_field)) {
			truncated_line(linenum, line);
			return;
		}
	}

	e = field_stop(time_field + 1);
	/* we have field TIME_FIELD in [time_field]..[e-1] */

	p = time_field;
	err = 0;
	f = 0;
	while (!SEP(*p)) {
		array[f] = str2ic(p);
		if (array[f] < 0) {
			array[f] = -1;
			err = 1;
		}
		if (++f == 5)
			break;
		SKIP_CHAR(p, '/');
	}

	if (unlikely(f < 5)) {
		parse_err++;
		return;
	}

	/* if we find at least one negative time, we count one error
	 * with a time equal to the total session time. This will
	 * emphasize quantum timing effects associated to known
	 * timeouts. Note that on some buggy machines, it is possible
	 * that the total time is negative, hence the reason to reset
	 * it.
	 */

	if (filter & FILT_GRAPH_TIMERS) {
		if (err) {
			if (array[4] < 0)
				array[4] = -1;
			t2 = insert_timer(&timers[0], tptr, array[4]);  // total time
			t2->count++;
		} else {
			int v;

			t2 = insert_timer(&timers[1], tptr, array[0]); t2->count++;  // req
			t2 = insert_timer(&timers[2], tptr, array[2]); t2->count++;  // conn
			t2 = insert_timer(&timers[3], tptr, array[3]); t2->count++;  // resp

			v = array[4] - array[0] - array[1] - array[2] - array[3]; // data time
			if (v < 0 && !(filter & FILT_QUIET))
				fprintf(stderr, "ERR: %s (%d %d %d %d %d => %d)\n",
					line, array[0], array[1], array[2], array[3], array[4], v);
			t2 = insert_timer(&timers[4], tptr, v); t2->count++;
			lines_out++;
		}
	} else { /* percentile */
		if (err) {
			if (array[4] < 0)
				array[4] = -1;
			t2 = insert_value(&timers[0], tptr, array[4]);  // total time
			t2->count++;
		} else {
			int v;

			t2 = insert_value(&timers[1], tptr, array[0]); t2->count++;  // req
			t2 = insert_value(&timers[2], tptr, array[2]); t2->count++;  // conn
			t2 = insert_value(&timers[3], tptr, array[3]); t2->count++;  // resp

			v = array[4] - array[0] - array[1] - array[2] - array[3]; // data time
			if (v < 0 && !(filter & FILT_QUIET))
				fprintf(stderr, "ERR: %s (%d %d %d %d %d => %d)\n",
					line, array[0], array[1], array[2], array[3], array[4], v);
			t2 = insert_value(&timers[4], tptr, v); t2->count++;
			lines_out++;
		}
	}
}


/*
 * Local variables:
 *  c-indent-level: 8
 *  c-basic-offset: 8
 * End:
 */