freebsd-dev/usr.sbin/syslogd/syslogd.c

1998 lines
45 KiB
C
Raw Normal View History

1994-05-26 05:23:31 +00:00
/*
* Copyright (c) 1983, 1988, 1993, 1994
* The Regents of the University of California. All rights reserved.
*
* 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.
* 3. All advertising materials mentioning features or use of this software
* must display the following acknowledgement:
* This product includes software developed by the University of
* California, Berkeley and its contributors.
* 4. Neither the name of the University nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE REGENTS 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 REGENTS 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.
*/
#ifndef lint
static const char copyright[] =
1994-05-26 05:23:31 +00:00
"@(#) Copyright (c) 1983, 1988, 1993, 1994\n\
The Regents of the University of California. All rights reserved.\n";
1997-10-20 12:55:49 +00:00
#endif /* not lint */
#ifndef lint
#if 0
1994-05-26 05:23:31 +00:00
static char sccsid[] = "@(#)syslogd.c 8.3 (Berkeley) 4/4/94";
1997-10-20 12:55:49 +00:00
#endif
static const char rcsid[] =
1999-08-28 01:35:59 +00:00
"$FreeBSD$";
1994-05-26 05:23:31 +00:00
#endif /* not lint */
/*
* syslogd -- log system messages
*
* This program implements a system log. It takes a series of lines.
* Each line may have a priority, signified as "<n>" as
* the first characters of the line. If this is
* not present, a default priority is used.
*
* To kill syslogd, send a signal 15 (terminate). A signal 1 (hup) will
* cause it to reread its configuration file.
*
* Defined Constants:
*
* MAXLINE -- the maximimum line length that can be handled.
* DEFUPRI -- the default priority for user messages
* DEFSPRI -- the default priority for kernel messages
*
* Author: Eric Allman
* extensive changes by Ralph Campbell
* more extensive changes by Eric Allman (again)
* Extension to log by program name as well as facility and priority
* by Peter da Silva.
* -u and -v by Harlan Stenn.
* Priority comparison code by Harlan Stenn.
1994-05-26 05:23:31 +00:00
*/
#define MAXLINE 1024 /* maximum line length */
#define MAXSVLINE 120 /* maximum saved line length */
#define DEFUPRI (LOG_USER|LOG_NOTICE)
#define DEFSPRI (LOG_KERN|LOG_CRIT)
#define TIMERINTVL 30 /* interval for checking flush, mark */
#define TTYMSGTIME 1 /* timed out passed to ttymsg */
1994-05-26 05:23:31 +00:00
#include <sys/param.h>
#include <sys/ioctl.h>
#include <sys/stat.h>
#include <sys/wait.h>
#include <sys/socket.h>
#include <sys/queue.h>
1994-05-26 05:23:31 +00:00
#include <sys/uio.h>
#include <sys/un.h>
#include <sys/time.h>
#include <sys/resource.h>
#include <sys/syslimits.h>
#include <paths.h>
1994-05-26 05:23:31 +00:00
#include <netinet/in.h>
#include <netdb.h>
#include <arpa/inet.h>
#include <ctype.h>
1997-10-20 12:55:49 +00:00
#include <err.h>
1994-05-26 05:23:31 +00:00
#include <errno.h>
#include <fcntl.h>
#include <regex.h>
1994-05-26 05:23:31 +00:00
#include <setjmp.h>
#include <signal.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sysexits.h>
1994-05-26 05:23:31 +00:00
#include <unistd.h>
#include <utmp.h>
#include "pathnames.h"
#define SYSLOG_NAMES
#include <sys/syslog.h>
const char *ConfFile = _PATH_LOGCONF;
const char *PidFile = _PATH_LOGPID;
const char ctty[] = _PATH_CONSOLE;
1994-05-26 05:23:31 +00:00
#define dprintf if (Debug) printf
#define MAXUNAMES 20 /* maximum number of user names */
#define MAXFUNIX 20
int nfunix = 1;
char *funixn[MAXFUNIX] = { _PATH_LOG };
int funix[MAXFUNIX];
1994-05-26 05:23:31 +00:00
/*
* Flags to logmsg().
*/
#define IGN_CONS 0x001 /* don't print on console */
#define SYNC_FILE 0x002 /* do fsync on file after printing */
#define ADDDATE 0x004 /* add a date to the message */
#define MARK 0x008 /* this message is a mark */
#define ISKERNEL 0x010 /* kernel generated message */
1994-05-26 05:23:31 +00:00
/*
* This structure represents the files that will have log
* copies printed.
*/
struct filed {
struct filed *f_next; /* next in linked list */
short f_type; /* entry type, see below */
short f_file; /* file descriptor */
time_t f_time; /* time this was last written */
u_char f_pmask[LOG_NFACILITIES+1]; /* priority mask */
u_char f_pcmp[LOG_NFACILITIES+1]; /* compare priority */
#define PRI_LT 0x1
#define PRI_EQ 0x2
#define PRI_GT 0x4
char *f_program; /* program this applies to */
1994-05-26 05:23:31 +00:00
union {
char f_uname[MAXUNAMES][UT_NAMESIZE+1];
struct {
char f_hname[MAXHOSTNAMELEN+1];
struct sockaddr_in f_addr;
} f_forw; /* forwarding address */
char f_fname[MAXPATHLEN];
struct {
char f_pname[MAXPATHLEN];
pid_t f_pid;
} f_pipe;
1994-05-26 05:23:31 +00:00
} f_un;
char f_prevline[MAXSVLINE]; /* last message logged */
char f_lasttime[16]; /* time of last occurrence */
char f_prevhost[MAXHOSTNAMELEN+1]; /* host from which recd. */
int f_prevpri; /* pri of f_prevline */
int f_prevlen; /* length of f_prevline */
int f_prevcount; /* repetition cnt of prevline */
int f_repeatcount; /* number of "repeated" msgs */
};
/*
* Queue of about-to-be dead processes we should watch out for.
*/
TAILQ_HEAD(stailhead, struct deadq_entry) deadq_head;
struct stailhead *deadq_headp;
struct deadq_entry {
pid_t dq_pid;
int dq_timeout;
TAILQ_ENTRY(struct deadq_entry) dq_entries;
};
/*
* The timeout to apply to processes waiting on the dead queue. Unit
* of measure is `mark intervals', i.e. 20 minutes by default.
* Processes on the dead queue will be terminated after that time.
*/
#define DQ_TIMO_INIT 2
typedef struct deadq_entry *dq_t;
/*
* Struct to hold records of network addresses that are allowed to log
* to us.
*/
struct allowedpeer {
int isnumeric;
u_short port;
union {
struct {
struct in_addr addr;
struct in_addr mask;
} numeric;
char *name;
} u;
#define a_addr u.numeric.addr
#define a_mask u.numeric.mask
#define a_name u.name
};
1994-05-26 05:23:31 +00:00
/*
* Intervals at which we flush out "message repeated" messages,
* in seconds after previous message is logged. After each flush,
* we move to the next interval until we reach the largest.
*/
int repeatinterval[] = { 30, 120, 600 }; /* # of secs before flush */
#define MAXREPEAT ((sizeof(repeatinterval) / sizeof(repeatinterval[0])) - 1)
#define REPEATTIME(f) ((f)->f_time + repeatinterval[(f)->f_repeatcount])
#define BACKOFF(f) { if (++(f)->f_repeatcount > MAXREPEAT) \
(f)->f_repeatcount = MAXREPEAT; \
}
/* values for f_type */
#define F_UNUSED 0 /* unused entry */
#define F_FILE 1 /* regular file */
#define F_TTY 2 /* terminal */
#define F_CONSOLE 3 /* console terminal */
#define F_FORW 4 /* remote machine */
#define F_USERS 5 /* list of users */
#define F_WALL 6 /* everyone logged on */
#define F_PIPE 7 /* pipe to program */
1994-05-26 05:23:31 +00:00
char *TypeNames[8] = {
1994-05-26 05:23:31 +00:00
"UNUSED", "FILE", "TTY", "CONSOLE",
"FORW", "USERS", "WALL", "PIPE"
1994-05-26 05:23:31 +00:00
};
struct filed *Files;
struct filed consfile;
int Debug; /* debug flag */
char LocalHostName[MAXHOSTNAMELEN+1]; /* our hostname */
char *LocalDomain; /* our local domain name */
int finet = -1; /* Internet datagram socket */
int fklog = -1; /* /dev/klog */
1994-05-26 05:23:31 +00:00
int LogPort; /* port number for INET connections */
int Initialized = 0; /* set when we have initialized ourselves */
int MarkInterval = 20 * 60; /* interval between marks in seconds */
int MarkSeq = 0; /* mark sequence number */
int SecureMode = 0; /* when true, receive only unix domain socks */
char bootfile[MAXLINE+1]; /* booted kernel file */
1994-05-26 05:23:31 +00:00
struct allowedpeer *AllowedPeers;
int NumAllowed = 0; /* # of AllowedPeer entries */
int UniquePriority = 0; /* Only log specified priority? */
int LogFacPri = 0; /* Put facility and priority in log message: */
/* 0=no, 1=numeric, 2=names */
int allowaddr __P((char *));
void cfline __P((char *, struct filed *, char *));
1994-05-26 05:23:31 +00:00
char *cvthname __P((struct sockaddr_in *));
Fix a serious bug in syslogd regarding the handling of pipes. The bug would cause syslogd to eventually kill innocent processes in the system over time (note: not `could' but `would'). Many thanks to my colleague Mirko for digging into the kernel structures and providing me with the debugging framework to find out about the nature of this bug (and to isolate that syslogd was the culprit) in a rather large set of distributed machines at client sites where this happened occasionally. Whenever a child process was no longer responsive, or when syslogd receives a SIGHUP so it closes all its logging file descriptors, for any descriptor that refers to a pipe syslogd enters the data about the old logging child process into a `dead queue', where it is being removed from (and the status of the dead kitten being fetched) upon receipt of a SIGCHLD. However, there's a high probability that the SIGCHLD already arrives before the child's data are actually entered into the dead queue inside the SIGHUP handler, so the SIGCHLD handler has nothing to fetch and remove and simply continues. Whenever this happens, the process'es data remain on the dead queue forever, and since domark() tried to get rid of totally unresponsive children by first sending a SIGTERM and later a SIGKILL, it was only a matter of time until the system had recycled enough PIDs so an innocent process got shot to death. Fix the race by masking SIGHUP and SIGCHLD from both handlers mutually. Add additional bandaids ``just in case'', i. e. don't enter a process into the dead queue if we can't signal it (this should only happen in case it is already dead by that time so we can fetch the status immediately instead of deferring this to the SIGCHLD handler); for the kill(2) inside domark(), check for an error status (/* Can't happen */ :) and remove it from the dead queue in this case (which if it would have been there in the first place would have reduced the problem to a statistically minimal likelihood so i certainly would never have noticed the bug at all :). Mirko also reviewed the fix in priciple (mutual blocking of both signals inside the handlers), but not the actual code. Reviewed by: Mirko Kaffka <mirko@interface-business.de> Approved by: jkh
2000-02-28 17:49:43 +00:00
void deadq_enter __P((pid_t, const char *));
int deadq_remove __P((pid_t));
1994-05-26 05:23:31 +00:00
int decode __P((const char *, CODE *));
void die __P((int));
void domark __P((int));
void fprintlog __P((struct filed *, int, char *));
void init __P((int));
1995-08-23 04:54:49 +00:00
void logerror __P((const char *));
1994-05-26 05:23:31 +00:00
void logmsg __P((int, char *, char *, int));
Fix a serious bug in syslogd regarding the handling of pipes. The bug would cause syslogd to eventually kill innocent processes in the system over time (note: not `could' but `would'). Many thanks to my colleague Mirko for digging into the kernel structures and providing me with the debugging framework to find out about the nature of this bug (and to isolate that syslogd was the culprit) in a rather large set of distributed machines at client sites where this happened occasionally. Whenever a child process was no longer responsive, or when syslogd receives a SIGHUP so it closes all its logging file descriptors, for any descriptor that refers to a pipe syslogd enters the data about the old logging child process into a `dead queue', where it is being removed from (and the status of the dead kitten being fetched) upon receipt of a SIGCHLD. However, there's a high probability that the SIGCHLD already arrives before the child's data are actually entered into the dead queue inside the SIGHUP handler, so the SIGCHLD handler has nothing to fetch and remove and simply continues. Whenever this happens, the process'es data remain on the dead queue forever, and since domark() tried to get rid of totally unresponsive children by first sending a SIGTERM and later a SIGKILL, it was only a matter of time until the system had recycled enough PIDs so an innocent process got shot to death. Fix the race by masking SIGHUP and SIGCHLD from both handlers mutually. Add additional bandaids ``just in case'', i. e. don't enter a process into the dead queue if we can't signal it (this should only happen in case it is already dead by that time so we can fetch the status immediately instead of deferring this to the SIGCHLD handler); for the kill(2) inside domark(), check for an error status (/* Can't happen */ :) and remove it from the dead queue in this case (which if it would have been there in the first place would have reduced the problem to a statistically minimal likelihood so i certainly would never have noticed the bug at all :). Mirko also reviewed the fix in priciple (mutual blocking of both signals inside the handlers), but not the actual code. Reviewed by: Mirko Kaffka <mirko@interface-business.de> Approved by: jkh
2000-02-28 17:49:43 +00:00
void log_deadchild __P((pid_t, int, const char *));
1994-05-26 05:23:31 +00:00
void printline __P((char *, char *));
void printsys __P((char *));
int p_open __P((char *, pid_t *));
void readklog __P((void));
1994-05-26 05:23:31 +00:00
void reapchild __P((int));
char *ttymsg __P((struct iovec *, int, char *, int));
1997-10-20 12:55:49 +00:00
static void usage __P((void));
int validate __P((struct sockaddr_in *, const char *));
1994-05-26 05:23:31 +00:00
void wallmsg __P((struct filed *, struct iovec *));
int waitdaemon __P((int, int, int));
void timedout __P((int));
1994-05-26 05:23:31 +00:00
int
main(argc, argv)
int argc;
char *argv[];
{
int ch, i, l;
1994-05-26 05:23:31 +00:00
struct sockaddr_un sunx, fromunix;
struct sockaddr_in sin, frominet;
FILE *fp;
char *p, *hname, line[MAXLINE + 1];
struct timeval tv, *tvp;
Fix a serious bug in syslogd regarding the handling of pipes. The bug would cause syslogd to eventually kill innocent processes in the system over time (note: not `could' but `would'). Many thanks to my colleague Mirko for digging into the kernel structures and providing me with the debugging framework to find out about the nature of this bug (and to isolate that syslogd was the culprit) in a rather large set of distributed machines at client sites where this happened occasionally. Whenever a child process was no longer responsive, or when syslogd receives a SIGHUP so it closes all its logging file descriptors, for any descriptor that refers to a pipe syslogd enters the data about the old logging child process into a `dead queue', where it is being removed from (and the status of the dead kitten being fetched) upon receipt of a SIGCHLD. However, there's a high probability that the SIGCHLD already arrives before the child's data are actually entered into the dead queue inside the SIGHUP handler, so the SIGCHLD handler has nothing to fetch and remove and simply continues. Whenever this happens, the process'es data remain on the dead queue forever, and since domark() tried to get rid of totally unresponsive children by first sending a SIGTERM and later a SIGKILL, it was only a matter of time until the system had recycled enough PIDs so an innocent process got shot to death. Fix the race by masking SIGHUP and SIGCHLD from both handlers mutually. Add additional bandaids ``just in case'', i. e. don't enter a process into the dead queue if we can't signal it (this should only happen in case it is already dead by that time so we can fetch the status immediately instead of deferring this to the SIGCHLD handler); for the kill(2) inside domark(), check for an error status (/* Can't happen */ :) and remove it from the dead queue in this case (which if it would have been there in the first place would have reduced the problem to a statistically minimal likelihood so i certainly would never have noticed the bug at all :). Mirko also reviewed the fix in priciple (mutual blocking of both signals inside the handlers), but not the actual code. Reviewed by: Mirko Kaffka <mirko@interface-business.de> Approved by: jkh
2000-02-28 17:49:43 +00:00
struct sigaction sact;
sigset_t mask;
pid_t ppid = 1;
socklen_t len;
1994-05-26 05:23:31 +00:00
while ((ch = getopt(argc, argv, "a:dl:f:m:p:suv")) != -1)
1994-05-26 05:23:31 +00:00
switch(ch) {
case 'd': /* debug */
Debug++;
break;
case 'a': /* allow specific network addresses only */
if (allowaddr(optarg) == -1)
usage();
break;
1994-05-26 05:23:31 +00:00
case 'f': /* configuration file */
ConfFile = optarg;
break;
case 'm': /* mark interval */
MarkInterval = atoi(optarg) * 60;
break;
case 'p': /* path */
funixn[0] = optarg;
1994-05-26 05:23:31 +00:00
break;
case 's': /* no network mode */
SecureMode++;
break;
case 'l':
if (nfunix < MAXFUNIX)
funixn[nfunix++] = optarg;
else
warnx("out of descriptors, ignoring %s",
optarg);
break;
case 'u': /* only log specified priority */
UniquePriority++;
break;
case 'v': /* log facility and priority */
LogFacPri++;
break;
1994-05-26 05:23:31 +00:00
case '?':
default:
usage();
}
if ((argc -= optind) != 0)
usage();
if (!Debug) {
ppid = waitdaemon(0, 0, 30);
if (ppid < 0)
err(1, "could not become daemon");
} else
1994-05-26 05:23:31 +00:00
setlinebuf(stdout);
if (NumAllowed)
endservent();
1994-05-26 05:23:31 +00:00
consfile.f_type = F_CONSOLE;
(void)strcpy(consfile.f_un.f_fname, ctty + sizeof _PATH_DEV - 1);
1994-05-26 05:23:31 +00:00
(void)gethostname(LocalHostName, sizeof(LocalHostName));
if ((p = strchr(LocalHostName, '.')) != NULL) {
*p++ = '\0';
LocalDomain = p;
} else
LocalDomain = "";
(void)strcpy(bootfile, getbootfile());
1994-05-26 05:23:31 +00:00
(void)signal(SIGTERM, die);
(void)signal(SIGINT, Debug ? die : SIG_IGN);
(void)signal(SIGQUIT, Debug ? die : SIG_IGN);
Fix a serious bug in syslogd regarding the handling of pipes. The bug would cause syslogd to eventually kill innocent processes in the system over time (note: not `could' but `would'). Many thanks to my colleague Mirko for digging into the kernel structures and providing me with the debugging framework to find out about the nature of this bug (and to isolate that syslogd was the culprit) in a rather large set of distributed machines at client sites where this happened occasionally. Whenever a child process was no longer responsive, or when syslogd receives a SIGHUP so it closes all its logging file descriptors, for any descriptor that refers to a pipe syslogd enters the data about the old logging child process into a `dead queue', where it is being removed from (and the status of the dead kitten being fetched) upon receipt of a SIGCHLD. However, there's a high probability that the SIGCHLD already arrives before the child's data are actually entered into the dead queue inside the SIGHUP handler, so the SIGCHLD handler has nothing to fetch and remove and simply continues. Whenever this happens, the process'es data remain on the dead queue forever, and since domark() tried to get rid of totally unresponsive children by first sending a SIGTERM and later a SIGKILL, it was only a matter of time until the system had recycled enough PIDs so an innocent process got shot to death. Fix the race by masking SIGHUP and SIGCHLD from both handlers mutually. Add additional bandaids ``just in case'', i. e. don't enter a process into the dead queue if we can't signal it (this should only happen in case it is already dead by that time so we can fetch the status immediately instead of deferring this to the SIGCHLD handler); for the kill(2) inside domark(), check for an error status (/* Can't happen */ :) and remove it from the dead queue in this case (which if it would have been there in the first place would have reduced the problem to a statistically minimal likelihood so i certainly would never have noticed the bug at all :). Mirko also reviewed the fix in priciple (mutual blocking of both signals inside the handlers), but not the actual code. Reviewed by: Mirko Kaffka <mirko@interface-business.de> Approved by: jkh
2000-02-28 17:49:43 +00:00
/*
* We don't want the SIGCHLD and SIGHUP handlers to interfere
* with each other; they are likely candidates for being called
* simultaneously (SIGHUP closes pipe descriptor, process dies,
* SIGCHLD happens).
*/
sigemptyset(&mask);
sigaddset(&mask, SIGHUP);
sact.sa_handler = reapchild;
sact.sa_mask = mask;
sact.sa_flags = SA_RESTART;
Fix a serious bug in syslogd regarding the handling of pipes. The bug would cause syslogd to eventually kill innocent processes in the system over time (note: not `could' but `would'). Many thanks to my colleague Mirko for digging into the kernel structures and providing me with the debugging framework to find out about the nature of this bug (and to isolate that syslogd was the culprit) in a rather large set of distributed machines at client sites where this happened occasionally. Whenever a child process was no longer responsive, or when syslogd receives a SIGHUP so it closes all its logging file descriptors, for any descriptor that refers to a pipe syslogd enters the data about the old logging child process into a `dead queue', where it is being removed from (and the status of the dead kitten being fetched) upon receipt of a SIGCHLD. However, there's a high probability that the SIGCHLD already arrives before the child's data are actually entered into the dead queue inside the SIGHUP handler, so the SIGCHLD handler has nothing to fetch and remove and simply continues. Whenever this happens, the process'es data remain on the dead queue forever, and since domark() tried to get rid of totally unresponsive children by first sending a SIGTERM and later a SIGKILL, it was only a matter of time until the system had recycled enough PIDs so an innocent process got shot to death. Fix the race by masking SIGHUP and SIGCHLD from both handlers mutually. Add additional bandaids ``just in case'', i. e. don't enter a process into the dead queue if we can't signal it (this should only happen in case it is already dead by that time so we can fetch the status immediately instead of deferring this to the SIGCHLD handler); for the kill(2) inside domark(), check for an error status (/* Can't happen */ :) and remove it from the dead queue in this case (which if it would have been there in the first place would have reduced the problem to a statistically minimal likelihood so i certainly would never have noticed the bug at all :). Mirko also reviewed the fix in priciple (mutual blocking of both signals inside the handlers), but not the actual code. Reviewed by: Mirko Kaffka <mirko@interface-business.de> Approved by: jkh
2000-02-28 17:49:43 +00:00
(void)sigaction(SIGCHLD, &sact, NULL);
1994-05-26 05:23:31 +00:00
(void)signal(SIGALRM, domark);
(void)signal(SIGPIPE, SIG_IGN); /* We'll catch EPIPE instead. */
1994-05-26 05:23:31 +00:00
(void)alarm(TIMERINTVL);
TAILQ_INIT(&deadq_head);
1994-05-26 05:23:31 +00:00
#ifndef SUN_LEN
#define SUN_LEN(unp) (strlen((unp)->sun_path) + 2)
#endif
for (i = 0; i < nfunix; i++) {
memset(&sunx, 0, sizeof(sunx));
sunx.sun_family = AF_UNIX;
(void)strncpy(sunx.sun_path, funixn[i], sizeof(sunx.sun_path));
funix[i] = socket(AF_UNIX, SOCK_DGRAM, 0);
if (funix[i] < 0 ||
bind(funix[i], (struct sockaddr *)&sunx,
SUN_LEN(&sunx)) < 0 ||
chmod(funixn[i], 0666) < 0) {
(void) snprintf(line, sizeof line,
"cannot create %s", funixn[i]);
logerror(line);
dprintf("cannot create %s (%d)\n", funixn[i], errno);
if (i == 0)
die(0);
}
}
if (SecureMode <= 1)
finet = socket(AF_INET, SOCK_DGRAM, 0);
if (finet >= 0) {
struct servent *sp;
sp = getservbyname("syslog", "udp");
if (sp == NULL) {
errno = 0;
logerror("syslog/udp: unknown service");
die(0);
}
memset(&sin, 0, sizeof(sin));
sin.sin_family = AF_INET;
sin.sin_port = LogPort = sp->s_port;
if (bind(finet, (struct sockaddr *)&sin, sizeof(sin)) < 0) {
logerror("bind");
if (!Debug)
die(0);
}
1994-05-26 05:23:31 +00:00
}
if (finet >= 0 && SecureMode) {
if (shutdown(finet, SHUT_RD) < 0) {
logerror("shutdown");
if (!Debug)
die(0);
}
}
if ((fklog = open(_PATH_KLOG, O_RDONLY, 0)) >= 0)
if (fcntl(fklog, F_SETFL, O_NONBLOCK) < 0)
fklog = -1;
if (fklog < 0)
1994-05-26 05:23:31 +00:00
dprintf("can't open %s (%d)\n", _PATH_KLOG, errno);
/* tuck my process id away */
fp = fopen(PidFile, "w");
if (fp != NULL) {
fprintf(fp, "%d\n", getpid());
(void) fclose(fp);
}
dprintf("off & running....\n");
init(0);
Fix a serious bug in syslogd regarding the handling of pipes. The bug would cause syslogd to eventually kill innocent processes in the system over time (note: not `could' but `would'). Many thanks to my colleague Mirko for digging into the kernel structures and providing me with the debugging framework to find out about the nature of this bug (and to isolate that syslogd was the culprit) in a rather large set of distributed machines at client sites where this happened occasionally. Whenever a child process was no longer responsive, or when syslogd receives a SIGHUP so it closes all its logging file descriptors, for any descriptor that refers to a pipe syslogd enters the data about the old logging child process into a `dead queue', where it is being removed from (and the status of the dead kitten being fetched) upon receipt of a SIGCHLD. However, there's a high probability that the SIGCHLD already arrives before the child's data are actually entered into the dead queue inside the SIGHUP handler, so the SIGCHLD handler has nothing to fetch and remove and simply continues. Whenever this happens, the process'es data remain on the dead queue forever, and since domark() tried to get rid of totally unresponsive children by first sending a SIGTERM and later a SIGKILL, it was only a matter of time until the system had recycled enough PIDs so an innocent process got shot to death. Fix the race by masking SIGHUP and SIGCHLD from both handlers mutually. Add additional bandaids ``just in case'', i. e. don't enter a process into the dead queue if we can't signal it (this should only happen in case it is already dead by that time so we can fetch the status immediately instead of deferring this to the SIGCHLD handler); for the kill(2) inside domark(), check for an error status (/* Can't happen */ :) and remove it from the dead queue in this case (which if it would have been there in the first place would have reduced the problem to a statistically minimal likelihood so i certainly would never have noticed the bug at all :). Mirko also reviewed the fix in priciple (mutual blocking of both signals inside the handlers), but not the actual code. Reviewed by: Mirko Kaffka <mirko@interface-business.de> Approved by: jkh
2000-02-28 17:49:43 +00:00
/* prevent SIGHUP and SIGCHLD handlers from running in parallel */
sigemptyset(&mask);
sigaddset(&mask, SIGCHLD);
sact.sa_handler = init;
sact.sa_mask = mask;
sact.sa_flags = SA_RESTART;
Fix a serious bug in syslogd regarding the handling of pipes. The bug would cause syslogd to eventually kill innocent processes in the system over time (note: not `could' but `would'). Many thanks to my colleague Mirko for digging into the kernel structures and providing me with the debugging framework to find out about the nature of this bug (and to isolate that syslogd was the culprit) in a rather large set of distributed machines at client sites where this happened occasionally. Whenever a child process was no longer responsive, or when syslogd receives a SIGHUP so it closes all its logging file descriptors, for any descriptor that refers to a pipe syslogd enters the data about the old logging child process into a `dead queue', where it is being removed from (and the status of the dead kitten being fetched) upon receipt of a SIGCHLD. However, there's a high probability that the SIGCHLD already arrives before the child's data are actually entered into the dead queue inside the SIGHUP handler, so the SIGCHLD handler has nothing to fetch and remove and simply continues. Whenever this happens, the process'es data remain on the dead queue forever, and since domark() tried to get rid of totally unresponsive children by first sending a SIGTERM and later a SIGKILL, it was only a matter of time until the system had recycled enough PIDs so an innocent process got shot to death. Fix the race by masking SIGHUP and SIGCHLD from both handlers mutually. Add additional bandaids ``just in case'', i. e. don't enter a process into the dead queue if we can't signal it (this should only happen in case it is already dead by that time so we can fetch the status immediately instead of deferring this to the SIGCHLD handler); for the kill(2) inside domark(), check for an error status (/* Can't happen */ :) and remove it from the dead queue in this case (which if it would have been there in the first place would have reduced the problem to a statistically minimal likelihood so i certainly would never have noticed the bug at all :). Mirko also reviewed the fix in priciple (mutual blocking of both signals inside the handlers), but not the actual code. Reviewed by: Mirko Kaffka <mirko@interface-business.de> Approved by: jkh
2000-02-28 17:49:43 +00:00
(void)sigaction(SIGHUP, &sact, NULL);
1994-05-26 05:23:31 +00:00
tvp = &tv;
tv.tv_sec = tv.tv_usec = 0;
1994-05-26 05:23:31 +00:00
for (;;) {
fd_set readfds;
int nfds = 0;
FD_ZERO(&readfds);
if (fklog != -1) {
FD_SET(fklog, &readfds);
if (fklog > nfds)
nfds = fklog;
}
if (finet != -1 && !SecureMode) {
FD_SET(finet, &readfds);
if (finet > nfds)
nfds = finet;
}
for (i = 0; i < nfunix; i++) {
if (funix[i] != -1) {
FD_SET(funix[i], &readfds);
if (funix[i] > nfds)
nfds = funix[i];
}
}
1994-05-26 05:23:31 +00:00
/*dprintf("readfds = %#x\n", readfds);*/
1998-07-02 19:35:40 +00:00
nfds = select(nfds+1, &readfds, (fd_set *)NULL,
(fd_set *)NULL, tvp);
if (nfds == 0) {
if (tvp) {
tvp = NULL;
if (ppid != 1)
kill(ppid, SIGALRM);
}
1994-05-26 05:23:31 +00:00
continue;
}
1994-05-26 05:23:31 +00:00
if (nfds < 0) {
if (errno != EINTR)
logerror("select");
continue;
}
/*dprintf("got a message (%d, %#x)\n", nfds, readfds);*/
if (fklog != -1 && FD_ISSET(fklog, &readfds))
readklog();
if (finet != -1 && FD_ISSET(finet, &readfds)) {
1994-05-26 05:23:31 +00:00
len = sizeof(frominet);
l = recvfrom(finet, line, MAXLINE, 0,
1994-05-26 05:23:31 +00:00
(struct sockaddr *)&frominet, &len);
if (l > 0) {
line[l] = '\0';
hname = cvthname(&frominet);
if (validate(&frominet, hname))
printline(hname, line);
} else if (l < 0 && errno != EINTR)
1994-05-26 05:23:31 +00:00
logerror("recvfrom inet");
1995-05-30 03:57:47 +00:00
}
for (i = 0; i < nfunix; i++) {
if (funix[i] != -1 && FD_ISSET(funix[i], &readfds)) {
len = sizeof(fromunix);
l = recvfrom(funix[i], line, MAXLINE, 0,
(struct sockaddr *)&fromunix, &len);
if (l > 0) {
line[l] = '\0';
printline(LocalHostName, line);
} else if (l < 0 && errno != EINTR)
logerror("recvfrom unix");
}
}
1994-05-26 05:23:31 +00:00
}
}
1997-10-20 12:55:49 +00:00
static void
1994-05-26 05:23:31 +00:00
usage()
{
fprintf(stderr, "%s\n%s\n%s\n",
"usage: syslogd [-dsuv] [-a allowed_peer] [-f config_file]",
" [-m mark_interval] [-p log_socket]",
" [-l log_socket]");
1994-05-26 05:23:31 +00:00
exit(1);
}
/*
* Take a raw input line, decode the message, and print the message
* on the appropriate log files.
*/
void
printline(hname, msg)
char *hname;
char *msg;
{
int c, pri;
char *p, *q, line[MAXLINE + 1];
/* test for special codes */
pri = DEFUPRI;
p = msg;
if (*p == '<') {
pri = 0;
while (isdigit(*++p))
pri = 10 * pri + (*p - '0');
if (*p == '>')
++p;
}
if (pri &~ (LOG_FACMASK|LOG_PRIMASK))
pri = DEFUPRI;
/* don't allow users to log kernel messages */
if (LOG_FAC(pri) == LOG_KERN)
pri = LOG_MAKEPRI(LOG_USER, LOG_PRI(pri));
q = line;
while ((c = (unsigned char)*p++) != '\0' &&
q < &line[sizeof(line) - 3]) {
if ((c & 0x80) && c < 0xA0) {
c &= 0x7F;
*q++ = 'M';
*q++ = '-';
}
if (isascii(c) && iscntrl(c)) {
1994-05-26 05:23:31 +00:00
if (c == '\n')
*q++ = ' ';
else if (c == '\t')
*q++ = '\t';
else {
*q++ = '^';
*q++ = c ^ 0100;
}
} else
1994-05-26 05:23:31 +00:00
*q++ = c;
}
1994-05-26 05:23:31 +00:00
*q = '\0';
logmsg(pri, line, hname, 0);
}
/*
* Read /dev/klog while data are available, split into lines.
1994-05-26 05:23:31 +00:00
*/
void
readklog()
1994-05-26 05:23:31 +00:00
{
char *p, *q, line[MAXLINE + 1];
int len, i;
1994-05-26 05:23:31 +00:00
len = 0;
for (;;) {
i = read(fklog, line + len, MAXLINE - 1 - len);
if (i > 0)
line[i + len] = '\0';
else if (i < 0 && errno != EINTR && errno != EAGAIN) {
logerror("klog");
fklog = -1;
break;
} else
break;
for (p = line; (q = strchr(p, '\n')) != NULL; p = q + 1) {
*q = '\0';
printsys(p);
}
len = strlen(p);
if (len >= MAXLINE - 1) {
printsys(p);
len = 0;
1994-05-26 05:23:31 +00:00
}
if (len > 0)
memmove(line, p, len + 1);
}
if (len > 0)
printsys(line);
}
/*
* Take a raw input line from /dev/klog, format similar to syslog().
*/
void
printsys(p)
char *p;
{
int pri, flags;
flags = ISKERNEL | SYNC_FILE | ADDDATE; /* fsync after write */
pri = DEFSPRI;
if (*p == '<') {
pri = 0;
while (isdigit(*++p))
pri = 10 * pri + (*p - '0');
if (*p == '>')
++p;
} else {
/* kernel printf's come out on console */
flags |= IGN_CONS;
1994-05-26 05:23:31 +00:00
}
if (pri &~ (LOG_FACMASK|LOG_PRIMASK))
pri = DEFSPRI;
logmsg(pri, p, LocalHostName, flags);
1994-05-26 05:23:31 +00:00
}
time_t now;
/*
* Log a message to the appropriate log files, users, etc. based on
* the priority.
*/
void
logmsg(pri, msg, from, flags)
int pri;
char *msg, *from;
int flags;
{
struct filed *f;
int i, fac, msglen, omask, prilev;
1994-05-26 05:23:31 +00:00
char *timestamp;
char prog[NAME_MAX+1];
char buf[MAXLINE+1];
1994-05-26 05:23:31 +00:00
dprintf("logmsg: pri %o, flags %x, from %s, msg %s\n",
pri, flags, from, msg);
omask = sigblock(sigmask(SIGHUP)|sigmask(SIGALRM));
/*
* Check to see if msg looks non-standard.
*/
msglen = strlen(msg);
if (msglen < 16 || msg[3] != ' ' || msg[6] != ' ' ||
msg[9] != ':' || msg[12] != ':' || msg[15] != ' ')
flags |= ADDDATE;
(void)time(&now);
if (flags & ADDDATE)
timestamp = ctime(&now) + 4;
else {
timestamp = msg;
msg += 16;
msglen -= 16;
}
/* skip leading blanks */
while(isspace(*msg)) {
msg++;
msglen--;
}
1994-05-26 05:23:31 +00:00
/* extract facility and priority level */
if (flags & MARK)
fac = LOG_NFACILITIES;
else
fac = LOG_FAC(pri);
prilev = LOG_PRI(pri);
/* extract program name */
for(i = 0; i < NAME_MAX; i++) {
if(!isalnum(msg[i]))
break;
prog[i] = msg[i];
}
prog[i] = 0;
/* add kernel prefix for kernel messages */
if (flags & ISKERNEL) {
snprintf(buf, sizeof(buf), "%s: %s", bootfile, msg);
msg = buf;
msglen = strlen(buf);
}
1994-05-26 05:23:31 +00:00
/* log the message to the particular outputs */
if (!Initialized) {
f = &consfile;
f->f_file = open(ctty, O_WRONLY, 0);
if (f->f_file >= 0) {
fprintlog(f, flags, msg);
(void)close(f->f_file);
}
(void)sigsetmask(omask);
return;
}
for (f = Files; f; f = f->f_next) {
/* skip messages that are incorrect priority */
if (!(((f->f_pcmp[fac] & PRI_EQ) && (f->f_pmask[fac] == prilev))
||((f->f_pcmp[fac] & PRI_LT) && (f->f_pmask[fac] < prilev))
||((f->f_pcmp[fac] & PRI_GT) && (f->f_pmask[fac] > prilev))
)
|| f->f_pmask[fac] == INTERNAL_NOPRI)
1994-05-26 05:23:31 +00:00
continue;
/* skip messages with the incorrect program name */
if(f->f_program)
if(strcmp(prog, f->f_program) != 0)
continue;
1994-05-26 05:23:31 +00:00
if (f->f_type == F_CONSOLE && (flags & IGN_CONS))
continue;
/* don't output marks to recently written files */
if ((flags & MARK) && (now - f->f_time) < MarkInterval / 2)
continue;
/*
* suppress duplicate lines to this file
*/
if ((flags & MARK) == 0 && msglen == f->f_prevlen &&
!strcmp(msg, f->f_prevline) &&
!strcasecmp(from, f->f_prevhost)) {
1994-05-26 05:23:31 +00:00
(void)strncpy(f->f_lasttime, timestamp, 15);
f->f_prevcount++;
dprintf("msg repeated %d times, %ld sec of %d\n",
1998-07-06 20:28:08 +00:00
f->f_prevcount, (long)(now - f->f_time),
1994-05-26 05:23:31 +00:00
repeatinterval[f->f_repeatcount]);
/*
* If domark would have logged this by now,
* flush it now (so we don't hold isolated messages),
* but back off so we'll flush less often
* in the future.
*/
if (now > REPEATTIME(f)) {
fprintlog(f, flags, (char *)NULL);
BACKOFF(f);
}
} else {
/* new line, save it */
if (f->f_prevcount)
fprintlog(f, 0, (char *)NULL);
f->f_repeatcount = 0;
f->f_prevpri = pri;
1994-05-26 05:23:31 +00:00
(void)strncpy(f->f_lasttime, timestamp, 15);
(void)strncpy(f->f_prevhost, from,
sizeof(f->f_prevhost)-1);
f->f_prevhost[sizeof(f->f_prevhost)-1] = '\0';
1994-05-26 05:23:31 +00:00
if (msglen < MAXSVLINE) {
f->f_prevlen = msglen;
(void)strcpy(f->f_prevline, msg);
fprintlog(f, flags, (char *)NULL);
} else {
f->f_prevline[0] = 0;
f->f_prevlen = 0;
fprintlog(f, flags, msg);
}
}
}
(void)sigsetmask(omask);
}
void
fprintlog(f, flags, msg)
struct filed *f;
int flags;
char *msg;
{
struct iovec iov[7];
1994-05-26 05:23:31 +00:00
struct iovec *v;
int l;
char line[MAXLINE + 1], repbuf[80], greetings[200];
char *msgret;
1994-05-26 05:23:31 +00:00
v = iov;
if (f->f_type == F_WALL) {
v->iov_base = greetings;
v->iov_len = snprintf(greetings, sizeof greetings,
1994-05-26 05:23:31 +00:00
"\r\n\7Message from syslogd@%s at %.24s ...\r\n",
f->f_prevhost, ctime(&now));
v++;
v->iov_base = "";
v->iov_len = 0;
v++;
} else {
v->iov_base = f->f_lasttime;
v->iov_len = 15;
v++;
v->iov_base = " ";
v->iov_len = 1;
v++;
}
if (LogFacPri) {
static char fp_buf[30]; /* Hollow laugh */
int fac = f->f_prevpri & LOG_FACMASK;
int pri = LOG_PRI(f->f_prevpri);
char *f_s = 0;
char f_n[5]; /* Hollow laugh */
char *p_s = 0;
char p_n[5]; /* Hollow laugh */
if (LogFacPri > 1) {
CODE *c;
for (c = facilitynames; c->c_name; c++) {
if (c->c_val == fac) {
f_s = c->c_name;
break;
}
}
for (c = prioritynames; c->c_name; c++) {
if (c->c_val == pri) {
p_s = c->c_name;
break;
}
}
}
if (!f_s) {
snprintf(f_n, sizeof f_n, "%d", LOG_FAC(fac));
f_s = f_n;
}
if (!p_s) {
snprintf(p_n, sizeof p_n, "%d", pri);
p_s = p_n;
}
snprintf(fp_buf, sizeof fp_buf, "<%s.%s> ", f_s, p_s);
v->iov_base = fp_buf;
v->iov_len = strlen(fp_buf);
} else {
v->iov_base="";
v->iov_len = 0;
}
v++;
1994-05-26 05:23:31 +00:00
v->iov_base = f->f_prevhost;
v->iov_len = strlen(v->iov_base);
v++;
v->iov_base = " ";
v->iov_len = 1;
v++;
if (msg) {
v->iov_base = msg;
v->iov_len = strlen(msg);
} else if (f->f_prevcount > 1) {
v->iov_base = repbuf;
v->iov_len = sprintf(repbuf, "last message repeated %d times",
f->f_prevcount);
} else {
v->iov_base = f->f_prevline;
v->iov_len = f->f_prevlen;
}
v++;
dprintf("Logging to %s", TypeNames[f->f_type]);
f->f_time = now;
switch (f->f_type) {
case F_UNUSED:
dprintf("\n");
break;
case F_FORW:
dprintf(" %s\n", f->f_un.f_forw.f_hname);
/* check for local vs remote messages */
if (strcasecmp(f->f_prevhost, LocalHostName))
l = snprintf(line, sizeof line - 1,
"<%d>%.15s Forwarded from %s: %s",
f->f_prevpri, iov[0].iov_base, f->f_prevhost,
iov[5].iov_base);
else
l = snprintf(line, sizeof line - 1, "<%d>%.15s %s",
f->f_prevpri, iov[0].iov_base, iov[5].iov_base);
1994-05-26 05:23:31 +00:00
if (l > MAXLINE)
l = MAXLINE;
if ((finet >= 0) &&
(sendto(finet, line, l, 0,
(struct sockaddr *)&f->f_un.f_forw.f_addr,
sizeof(f->f_un.f_forw.f_addr)) != l)) {
1994-05-26 05:23:31 +00:00
int e = errno;
(void)close(f->f_file);
f->f_type = F_UNUSED;
errno = e;
logerror("sendto");
}
break;
case F_FILE:
dprintf(" %s\n", f->f_un.f_fname);
v->iov_base = "\n";
v->iov_len = 1;
if (writev(f->f_file, iov, 7) < 0) {
1994-05-26 05:23:31 +00:00
int e = errno;
(void)close(f->f_file);
f->f_type = F_UNUSED;
errno = e;
logerror(f->f_un.f_fname);
1994-05-26 05:23:31 +00:00
} else if (flags & SYNC_FILE)
(void)fsync(f->f_file);
break;
case F_PIPE:
dprintf(" %s\n", f->f_un.f_pipe.f_pname);
v->iov_base = "\n";
v->iov_len = 1;
if (f->f_un.f_pipe.f_pid == 0) {
if ((f->f_file = p_open(f->f_un.f_pipe.f_pname,
&f->f_un.f_pipe.f_pid)) < 0) {
f->f_type = F_UNUSED;
logerror(f->f_un.f_pipe.f_pname);
break;
}
}
if (writev(f->f_file, iov, 7) < 0) {
int e = errno;
(void)close(f->f_file);
if (f->f_un.f_pipe.f_pid > 0)
Fix a serious bug in syslogd regarding the handling of pipes. The bug would cause syslogd to eventually kill innocent processes in the system over time (note: not `could' but `would'). Many thanks to my colleague Mirko for digging into the kernel structures and providing me with the debugging framework to find out about the nature of this bug (and to isolate that syslogd was the culprit) in a rather large set of distributed machines at client sites where this happened occasionally. Whenever a child process was no longer responsive, or when syslogd receives a SIGHUP so it closes all its logging file descriptors, for any descriptor that refers to a pipe syslogd enters the data about the old logging child process into a `dead queue', where it is being removed from (and the status of the dead kitten being fetched) upon receipt of a SIGCHLD. However, there's a high probability that the SIGCHLD already arrives before the child's data are actually entered into the dead queue inside the SIGHUP handler, so the SIGCHLD handler has nothing to fetch and remove and simply continues. Whenever this happens, the process'es data remain on the dead queue forever, and since domark() tried to get rid of totally unresponsive children by first sending a SIGTERM and later a SIGKILL, it was only a matter of time until the system had recycled enough PIDs so an innocent process got shot to death. Fix the race by masking SIGHUP and SIGCHLD from both handlers mutually. Add additional bandaids ``just in case'', i. e. don't enter a process into the dead queue if we can't signal it (this should only happen in case it is already dead by that time so we can fetch the status immediately instead of deferring this to the SIGCHLD handler); for the kill(2) inside domark(), check for an error status (/* Can't happen */ :) and remove it from the dead queue in this case (which if it would have been there in the first place would have reduced the problem to a statistically minimal likelihood so i certainly would never have noticed the bug at all :). Mirko also reviewed the fix in priciple (mutual blocking of both signals inside the handlers), but not the actual code. Reviewed by: Mirko Kaffka <mirko@interface-business.de> Approved by: jkh
2000-02-28 17:49:43 +00:00
deadq_enter(f->f_un.f_pipe.f_pid,
f->f_un.f_pipe.f_pname);
f->f_un.f_pipe.f_pid = 0;
errno = e;
logerror(f->f_un.f_pipe.f_pname);
}
break;
case F_CONSOLE:
if (flags & IGN_CONS) {
dprintf(" (ignored)\n");
break;
}
/* FALLTHROUGH */
case F_TTY:
dprintf(" %s%s\n", _PATH_DEV, f->f_un.f_fname);
v->iov_base = "\r\n";
v->iov_len = 2;
errno = 0; /* ttymsg() only sometimes returns an errno */
if ((msgret = ttymsg(iov, 7, f->f_un.f_fname, 10))) {
f->f_type = F_UNUSED;
logerror(msgret);
}
break;
1994-05-26 05:23:31 +00:00
case F_USERS:
case F_WALL:
dprintf("\n");
v->iov_base = "\r\n";
v->iov_len = 2;
wallmsg(f, iov);
break;
}
f->f_prevcount = 0;
}
/*
* WALLMSG -- Write a message to the world at large
*
* Write the specified message to either the entire
* world, or a list of approved users.
*/
void
wallmsg(f, iov)
struct filed *f;
struct iovec *iov;
{
static int reenter; /* avoid calling ourselves */
FILE *uf;
struct utmp ut;
int i;
char *p;
char line[sizeof(ut.ut_line) + 1];
if (reenter++)
return;
if ((uf = fopen(_PATH_UTMP, "r")) == NULL) {
logerror(_PATH_UTMP);
reenter = 0;
return;
}
/* NOSTRICT */
while (fread((char *)&ut, sizeof(ut), 1, uf) == 1) {
if (ut.ut_name[0] == '\0')
continue;
strncpy(line, ut.ut_line, sizeof(ut.ut_line));
line[sizeof(ut.ut_line)] = '\0';
if (f->f_type == F_WALL) {
if ((p = ttymsg(iov, 7, line, TTYMSGTIME)) != NULL) {
1994-05-26 05:23:31 +00:00
errno = 0; /* already in msg */
logerror(p);
}
continue;
}
/* should we send the message to this user? */
for (i = 0; i < MAXUNAMES; i++) {
if (!f->f_un.f_uname[i][0])
break;
if (!strncmp(f->f_un.f_uname[i], ut.ut_name,
UT_NAMESIZE)) {
if ((p = ttymsg(iov, 7, line, TTYMSGTIME))
!= NULL) {
1994-05-26 05:23:31 +00:00
errno = 0; /* already in msg */
logerror(p);
}
break;
}
}
}
(void)fclose(uf);
reenter = 0;
}
void
reapchild(signo)
int signo;
{
Fix a serious bug in syslogd regarding the handling of pipes. The bug would cause syslogd to eventually kill innocent processes in the system over time (note: not `could' but `would'). Many thanks to my colleague Mirko for digging into the kernel structures and providing me with the debugging framework to find out about the nature of this bug (and to isolate that syslogd was the culprit) in a rather large set of distributed machines at client sites where this happened occasionally. Whenever a child process was no longer responsive, or when syslogd receives a SIGHUP so it closes all its logging file descriptors, for any descriptor that refers to a pipe syslogd enters the data about the old logging child process into a `dead queue', where it is being removed from (and the status of the dead kitten being fetched) upon receipt of a SIGCHLD. However, there's a high probability that the SIGCHLD already arrives before the child's data are actually entered into the dead queue inside the SIGHUP handler, so the SIGCHLD handler has nothing to fetch and remove and simply continues. Whenever this happens, the process'es data remain on the dead queue forever, and since domark() tried to get rid of totally unresponsive children by first sending a SIGTERM and later a SIGKILL, it was only a matter of time until the system had recycled enough PIDs so an innocent process got shot to death. Fix the race by masking SIGHUP and SIGCHLD from both handlers mutually. Add additional bandaids ``just in case'', i. e. don't enter a process into the dead queue if we can't signal it (this should only happen in case it is already dead by that time so we can fetch the status immediately instead of deferring this to the SIGCHLD handler); for the kill(2) inside domark(), check for an error status (/* Can't happen */ :) and remove it from the dead queue in this case (which if it would have been there in the first place would have reduced the problem to a statistically minimal likelihood so i certainly would never have noticed the bug at all :). Mirko also reviewed the fix in priciple (mutual blocking of both signals inside the handlers), but not the actual code. Reviewed by: Mirko Kaffka <mirko@interface-business.de> Approved by: jkh
2000-02-28 17:49:43 +00:00
int status;
pid_t pid;
struct filed *f;
while ((pid = wait3(&status, WNOHANG, (struct rusage *)NULL)) > 0) {
if (!Initialized)
/* Don't tell while we are initting. */
continue;
1994-05-26 05:23:31 +00:00
/* First, look if it's a process from the dead queue. */
Fix a serious bug in syslogd regarding the handling of pipes. The bug would cause syslogd to eventually kill innocent processes in the system over time (note: not `could' but `would'). Many thanks to my colleague Mirko for digging into the kernel structures and providing me with the debugging framework to find out about the nature of this bug (and to isolate that syslogd was the culprit) in a rather large set of distributed machines at client sites where this happened occasionally. Whenever a child process was no longer responsive, or when syslogd receives a SIGHUP so it closes all its logging file descriptors, for any descriptor that refers to a pipe syslogd enters the data about the old logging child process into a `dead queue', where it is being removed from (and the status of the dead kitten being fetched) upon receipt of a SIGCHLD. However, there's a high probability that the SIGCHLD already arrives before the child's data are actually entered into the dead queue inside the SIGHUP handler, so the SIGCHLD handler has nothing to fetch and remove and simply continues. Whenever this happens, the process'es data remain on the dead queue forever, and since domark() tried to get rid of totally unresponsive children by first sending a SIGTERM and later a SIGKILL, it was only a matter of time until the system had recycled enough PIDs so an innocent process got shot to death. Fix the race by masking SIGHUP and SIGCHLD from both handlers mutually. Add additional bandaids ``just in case'', i. e. don't enter a process into the dead queue if we can't signal it (this should only happen in case it is already dead by that time so we can fetch the status immediately instead of deferring this to the SIGCHLD handler); for the kill(2) inside domark(), check for an error status (/* Can't happen */ :) and remove it from the dead queue in this case (which if it would have been there in the first place would have reduced the problem to a statistically minimal likelihood so i certainly would never have noticed the bug at all :). Mirko also reviewed the fix in priciple (mutual blocking of both signals inside the handlers), but not the actual code. Reviewed by: Mirko Kaffka <mirko@interface-business.de> Approved by: jkh
2000-02-28 17:49:43 +00:00
if (deadq_remove(pid))
goto oncemore;
/* Now, look in list of active processes. */
for (f = Files; f; f = f->f_next)
if (f->f_type == F_PIPE &&
f->f_un.f_pipe.f_pid == pid) {
(void)close(f->f_file);
f->f_un.f_pipe.f_pid = 0;
Fix a serious bug in syslogd regarding the handling of pipes. The bug would cause syslogd to eventually kill innocent processes in the system over time (note: not `could' but `would'). Many thanks to my colleague Mirko for digging into the kernel structures and providing me with the debugging framework to find out about the nature of this bug (and to isolate that syslogd was the culprit) in a rather large set of distributed machines at client sites where this happened occasionally. Whenever a child process was no longer responsive, or when syslogd receives a SIGHUP so it closes all its logging file descriptors, for any descriptor that refers to a pipe syslogd enters the data about the old logging child process into a `dead queue', where it is being removed from (and the status of the dead kitten being fetched) upon receipt of a SIGCHLD. However, there's a high probability that the SIGCHLD already arrives before the child's data are actually entered into the dead queue inside the SIGHUP handler, so the SIGCHLD handler has nothing to fetch and remove and simply continues. Whenever this happens, the process'es data remain on the dead queue forever, and since domark() tried to get rid of totally unresponsive children by first sending a SIGTERM and later a SIGKILL, it was only a matter of time until the system had recycled enough PIDs so an innocent process got shot to death. Fix the race by masking SIGHUP and SIGCHLD from both handlers mutually. Add additional bandaids ``just in case'', i. e. don't enter a process into the dead queue if we can't signal it (this should only happen in case it is already dead by that time so we can fetch the status immediately instead of deferring this to the SIGCHLD handler); for the kill(2) inside domark(), check for an error status (/* Can't happen */ :) and remove it from the dead queue in this case (which if it would have been there in the first place would have reduced the problem to a statistically minimal likelihood so i certainly would never have noticed the bug at all :). Mirko also reviewed the fix in priciple (mutual blocking of both signals inside the handlers), but not the actual code. Reviewed by: Mirko Kaffka <mirko@interface-business.de> Approved by: jkh
2000-02-28 17:49:43 +00:00
log_deadchild(pid, status,
f->f_un.f_pipe.f_pname);
break;
}
oncemore:
continue;
}
1994-05-26 05:23:31 +00:00
}
/*
* Return a printable representation of a host address.
*/
char *
cvthname(f)
struct sockaddr_in *f;
{
struct hostent *hp;
sigset_t omask, nmask;
1994-05-26 05:23:31 +00:00
char *p;
dprintf("cvthname(%s)\n", inet_ntoa(f->sin_addr));
if (f->sin_family != AF_INET) {
dprintf("Malformed from address\n");
return ("???");
}
sigemptyset(&nmask);
sigaddset(&nmask, SIGHUP);
sigprocmask(SIG_BLOCK, &nmask, &omask);
1994-05-26 05:23:31 +00:00
hp = gethostbyaddr((char *)&f->sin_addr,
sizeof(struct in_addr), f->sin_family);
sigprocmask(SIG_SETMASK, &omask, NULL);
1994-05-26 05:23:31 +00:00
if (hp == 0) {
dprintf("Host name for your address (%s) unknown\n",
inet_ntoa(f->sin_addr));
return (inet_ntoa(f->sin_addr));
}
if ((p = strchr(hp->h_name, '.')) &&
strcasecmp(p + 1, LocalDomain) == 0)
1994-05-26 05:23:31 +00:00
*p = '\0';
return (hp->h_name);
}
void
domark(signo)
int signo;
{
struct filed *f;
dq_t q;
1994-05-26 05:23:31 +00:00
now = time((time_t *)NULL);
MarkSeq += TIMERINTVL;
if (MarkSeq >= MarkInterval) {
logmsg(LOG_INFO, "-- MARK --", LocalHostName, ADDDATE|MARK);
MarkSeq = 0;
}
for (f = Files; f; f = f->f_next) {
if (f->f_prevcount && now >= REPEATTIME(f)) {
dprintf("flush %s: repeated %d times, %d sec.\n",
TypeNames[f->f_type], f->f_prevcount,
repeatinterval[f->f_repeatcount]);
fprintlog(f, 0, (char *)NULL);
BACKOFF(f);
}
}
/* Walk the dead queue, and see if we should signal somebody. */
for (q = TAILQ_FIRST(&deadq_head); q != NULL; q = TAILQ_NEXT(q, dq_entries))
switch (q->dq_timeout) {
case 0:
/* Already signalled once, try harder now. */
Fix a serious bug in syslogd regarding the handling of pipes. The bug would cause syslogd to eventually kill innocent processes in the system over time (note: not `could' but `would'). Many thanks to my colleague Mirko for digging into the kernel structures and providing me with the debugging framework to find out about the nature of this bug (and to isolate that syslogd was the culprit) in a rather large set of distributed machines at client sites where this happened occasionally. Whenever a child process was no longer responsive, or when syslogd receives a SIGHUP so it closes all its logging file descriptors, for any descriptor that refers to a pipe syslogd enters the data about the old logging child process into a `dead queue', where it is being removed from (and the status of the dead kitten being fetched) upon receipt of a SIGCHLD. However, there's a high probability that the SIGCHLD already arrives before the child's data are actually entered into the dead queue inside the SIGHUP handler, so the SIGCHLD handler has nothing to fetch and remove and simply continues. Whenever this happens, the process'es data remain on the dead queue forever, and since domark() tried to get rid of totally unresponsive children by first sending a SIGTERM and later a SIGKILL, it was only a matter of time until the system had recycled enough PIDs so an innocent process got shot to death. Fix the race by masking SIGHUP and SIGCHLD from both handlers mutually. Add additional bandaids ``just in case'', i. e. don't enter a process into the dead queue if we can't signal it (this should only happen in case it is already dead by that time so we can fetch the status immediately instead of deferring this to the SIGCHLD handler); for the kill(2) inside domark(), check for an error status (/* Can't happen */ :) and remove it from the dead queue in this case (which if it would have been there in the first place would have reduced the problem to a statistically minimal likelihood so i certainly would never have noticed the bug at all :). Mirko also reviewed the fix in priciple (mutual blocking of both signals inside the handlers), but not the actual code. Reviewed by: Mirko Kaffka <mirko@interface-business.de> Approved by: jkh
2000-02-28 17:49:43 +00:00
if (kill(q->dq_pid, SIGKILL) != 0)
(void)deadq_remove(q->dq_pid);
break;
case 1:
/*
* Timed out on dead queue, send terminate
* signal. Note that we leave the removal
* from the dead queue to reapchild(), which
Fix a serious bug in syslogd regarding the handling of pipes. The bug would cause syslogd to eventually kill innocent processes in the system over time (note: not `could' but `would'). Many thanks to my colleague Mirko for digging into the kernel structures and providing me with the debugging framework to find out about the nature of this bug (and to isolate that syslogd was the culprit) in a rather large set of distributed machines at client sites where this happened occasionally. Whenever a child process was no longer responsive, or when syslogd receives a SIGHUP so it closes all its logging file descriptors, for any descriptor that refers to a pipe syslogd enters the data about the old logging child process into a `dead queue', where it is being removed from (and the status of the dead kitten being fetched) upon receipt of a SIGCHLD. However, there's a high probability that the SIGCHLD already arrives before the child's data are actually entered into the dead queue inside the SIGHUP handler, so the SIGCHLD handler has nothing to fetch and remove and simply continues. Whenever this happens, the process'es data remain on the dead queue forever, and since domark() tried to get rid of totally unresponsive children by first sending a SIGTERM and later a SIGKILL, it was only a matter of time until the system had recycled enough PIDs so an innocent process got shot to death. Fix the race by masking SIGHUP and SIGCHLD from both handlers mutually. Add additional bandaids ``just in case'', i. e. don't enter a process into the dead queue if we can't signal it (this should only happen in case it is already dead by that time so we can fetch the status immediately instead of deferring this to the SIGCHLD handler); for the kill(2) inside domark(), check for an error status (/* Can't happen */ :) and remove it from the dead queue in this case (which if it would have been there in the first place would have reduced the problem to a statistically minimal likelihood so i certainly would never have noticed the bug at all :). Mirko also reviewed the fix in priciple (mutual blocking of both signals inside the handlers), but not the actual code. Reviewed by: Mirko Kaffka <mirko@interface-business.de> Approved by: jkh
2000-02-28 17:49:43 +00:00
* will also log the event (unless the process
* didn't even really exist, in case we simply
* drop it from the dead queue).
*/
Fix a serious bug in syslogd regarding the handling of pipes. The bug would cause syslogd to eventually kill innocent processes in the system over time (note: not `could' but `would'). Many thanks to my colleague Mirko for digging into the kernel structures and providing me with the debugging framework to find out about the nature of this bug (and to isolate that syslogd was the culprit) in a rather large set of distributed machines at client sites where this happened occasionally. Whenever a child process was no longer responsive, or when syslogd receives a SIGHUP so it closes all its logging file descriptors, for any descriptor that refers to a pipe syslogd enters the data about the old logging child process into a `dead queue', where it is being removed from (and the status of the dead kitten being fetched) upon receipt of a SIGCHLD. However, there's a high probability that the SIGCHLD already arrives before the child's data are actually entered into the dead queue inside the SIGHUP handler, so the SIGCHLD handler has nothing to fetch and remove and simply continues. Whenever this happens, the process'es data remain on the dead queue forever, and since domark() tried to get rid of totally unresponsive children by first sending a SIGTERM and later a SIGKILL, it was only a matter of time until the system had recycled enough PIDs so an innocent process got shot to death. Fix the race by masking SIGHUP and SIGCHLD from both handlers mutually. Add additional bandaids ``just in case'', i. e. don't enter a process into the dead queue if we can't signal it (this should only happen in case it is already dead by that time so we can fetch the status immediately instead of deferring this to the SIGCHLD handler); for the kill(2) inside domark(), check for an error status (/* Can't happen */ :) and remove it from the dead queue in this case (which if it would have been there in the first place would have reduced the problem to a statistically minimal likelihood so i certainly would never have noticed the bug at all :). Mirko also reviewed the fix in priciple (mutual blocking of both signals inside the handlers), but not the actual code. Reviewed by: Mirko Kaffka <mirko@interface-business.de> Approved by: jkh
2000-02-28 17:49:43 +00:00
if (kill(q->dq_pid, SIGTERM) != 0)
(void)deadq_remove(q->dq_pid);
/* FALLTHROUGH */
default:
q->dq_timeout--;
}
1994-05-26 05:23:31 +00:00
(void)alarm(TIMERINTVL);
}
/*
* Print syslogd errors some place.
*/
void
logerror(type)
1995-08-23 04:54:49 +00:00
const char *type;
1994-05-26 05:23:31 +00:00
{
char buf[512];
1994-05-26 05:23:31 +00:00
if (errno)
(void)snprintf(buf,
sizeof buf, "syslogd: %s: %s", type, strerror(errno));
1994-05-26 05:23:31 +00:00
else
(void)snprintf(buf, sizeof buf, "syslogd: %s", type);
1994-05-26 05:23:31 +00:00
errno = 0;
dprintf("%s\n", buf);
logmsg(LOG_SYSLOG|LOG_ERR, buf, LocalHostName, ADDDATE);
}
void
die(signo)
int signo;
{
struct filed *f;
int was_initialized;
1994-05-26 05:23:31 +00:00
char buf[100];
int i;
1994-05-26 05:23:31 +00:00
was_initialized = Initialized;
Initialized = 0; /* Don't log SIGCHLDs. */
1994-05-26 05:23:31 +00:00
for (f = Files; f != NULL; f = f->f_next) {
/* flush any pending output */
if (f->f_prevcount)
fprintlog(f, 0, (char *)NULL);
if (f->f_type == F_PIPE)
(void)close(f->f_file);
1994-05-26 05:23:31 +00:00
}
Initialized = was_initialized;
1994-05-26 05:23:31 +00:00
if (signo) {
dprintf("syslogd: exiting on signal %d\n", signo);
(void)sprintf(buf, "exiting on signal %d", signo);
errno = 0;
logerror(buf);
}
for (i = 0; i < nfunix; i++)
if (funixn[i] && funix[i] != -1)
(void)unlink(funixn[i]);
exit(1);
1994-05-26 05:23:31 +00:00
}
/*
* INIT -- Initialize syslogd from configuration table
*/
void
init(signo)
int signo;
{
int i;
FILE *cf;
struct filed *f, *next, **nextp;
char *p;
char cline[LINE_MAX];
char prog[NAME_MAX+1];
1994-05-26 05:23:31 +00:00
dprintf("init\n");
/*
* Close all open log files.
*/
Initialized = 0;
for (f = Files; f != NULL; f = next) {
/* flush any pending output */
if (f->f_prevcount)
fprintlog(f, 0, (char *)NULL);
switch (f->f_type) {
case F_FILE:
case F_FORW:
case F_CONSOLE:
case F_TTY:
(void)close(f->f_file);
break;
case F_PIPE:
(void)close(f->f_file);
if (f->f_un.f_pipe.f_pid > 0)
Fix a serious bug in syslogd regarding the handling of pipes. The bug would cause syslogd to eventually kill innocent processes in the system over time (note: not `could' but `would'). Many thanks to my colleague Mirko for digging into the kernel structures and providing me with the debugging framework to find out about the nature of this bug (and to isolate that syslogd was the culprit) in a rather large set of distributed machines at client sites where this happened occasionally. Whenever a child process was no longer responsive, or when syslogd receives a SIGHUP so it closes all its logging file descriptors, for any descriptor that refers to a pipe syslogd enters the data about the old logging child process into a `dead queue', where it is being removed from (and the status of the dead kitten being fetched) upon receipt of a SIGCHLD. However, there's a high probability that the SIGCHLD already arrives before the child's data are actually entered into the dead queue inside the SIGHUP handler, so the SIGCHLD handler has nothing to fetch and remove and simply continues. Whenever this happens, the process'es data remain on the dead queue forever, and since domark() tried to get rid of totally unresponsive children by first sending a SIGTERM and later a SIGKILL, it was only a matter of time until the system had recycled enough PIDs so an innocent process got shot to death. Fix the race by masking SIGHUP and SIGCHLD from both handlers mutually. Add additional bandaids ``just in case'', i. e. don't enter a process into the dead queue if we can't signal it (this should only happen in case it is already dead by that time so we can fetch the status immediately instead of deferring this to the SIGCHLD handler); for the kill(2) inside domark(), check for an error status (/* Can't happen */ :) and remove it from the dead queue in this case (which if it would have been there in the first place would have reduced the problem to a statistically minimal likelihood so i certainly would never have noticed the bug at all :). Mirko also reviewed the fix in priciple (mutual blocking of both signals inside the handlers), but not the actual code. Reviewed by: Mirko Kaffka <mirko@interface-business.de> Approved by: jkh
2000-02-28 17:49:43 +00:00
deadq_enter(f->f_un.f_pipe.f_pid,
f->f_un.f_pipe.f_pname);
f->f_un.f_pipe.f_pid = 0;
break;
1994-05-26 05:23:31 +00:00
}
next = f->f_next;
if(f->f_program) free(f->f_program);
1994-05-26 05:23:31 +00:00
free((char *)f);
}
Files = NULL;
nextp = &Files;
/* open the configuration file */
if ((cf = fopen(ConfFile, "r")) == NULL) {
dprintf("cannot open %s\n", ConfFile);
*nextp = (struct filed *)calloc(1, sizeof(*f));
cfline("*.ERR\t/dev/console", *nextp, "*");
1994-05-26 05:23:31 +00:00
(*nextp)->f_next = (struct filed *)calloc(1, sizeof(*f));
cfline("*.PANIC\t*", (*nextp)->f_next, "*");
1994-05-26 05:23:31 +00:00
Initialized = 1;
return;
}
/*
* Foreach line in the conf table, open that file.
*/
f = NULL;
strcpy(prog, "*");
1994-05-26 05:23:31 +00:00
while (fgets(cline, sizeof(cline), cf) != NULL) {
/*
* check for end-of-section, comments, strip off trailing
* spaces and newline character. #!prog is treated specially:
* following lines apply only to that program.
1994-05-26 05:23:31 +00:00
*/
for (p = cline; isspace(*p); ++p)
continue;
if (*p == 0)
continue;
if(*p == '#') {
p++;
if(*p!='!')
continue;
}
if(*p=='!') {
p++;
while(isspace(*p)) p++;
if((!*p) || (*p == '*')) {
strcpy(prog, "*");
continue;
}
for(i = 0; i < NAME_MAX; i++) {
if(!isalnum(p[i]))
break;
prog[i] = p[i];
}
prog[i] = 0;
1994-05-26 05:23:31 +00:00
continue;
}
1994-05-26 05:23:31 +00:00
for (p = strchr(cline, '\0'); isspace(*--p);)
continue;
*++p = '\0';
f = (struct filed *)calloc(1, sizeof(*f));
*nextp = f;
nextp = &f->f_next;
cfline(cline, f, prog);
1994-05-26 05:23:31 +00:00
}
/* close the configuration file */
(void)fclose(cf);
Initialized = 1;
if (Debug) {
for (f = Files; f; f = f->f_next) {
for (i = 0; i <= LOG_NFACILITIES; i++)
if (f->f_pmask[i] == INTERNAL_NOPRI)
printf("X ");
else
printf("%d ", f->f_pmask[i]);
printf("%s: ", TypeNames[f->f_type]);
switch (f->f_type) {
case F_FILE:
printf("%s", f->f_un.f_fname);
break;
case F_CONSOLE:
case F_TTY:
printf("%s%s", _PATH_DEV, f->f_un.f_fname);
break;
1994-05-26 05:23:31 +00:00
case F_FORW:
printf("%s", f->f_un.f_forw.f_hname);
break;
case F_PIPE:
printf("%s", f->f_un.f_pipe.f_pname);
break;
1994-05-26 05:23:31 +00:00
case F_USERS:
for (i = 0; i < MAXUNAMES && *f->f_un.f_uname[i]; i++)
printf("%s, ", f->f_un.f_uname[i]);
break;
}
if(f->f_program) {
printf(" (%s)", f->f_program);
}
1994-05-26 05:23:31 +00:00
printf("\n");
}
}
logmsg(LOG_SYSLOG|LOG_INFO, "syslogd: restart", LocalHostName, ADDDATE);
dprintf("syslogd: restarted\n");
}
/*
* Crack a configuration file line
*/
void
cfline(line, f, prog)
1994-05-26 05:23:31 +00:00
char *line;
struct filed *f;
char *prog;
1994-05-26 05:23:31 +00:00
{
struct hostent *hp;
int i, pri;
char *bp, *p, *q;
char buf[MAXLINE], ebuf[100];
dprintf("cfline(\"%s\", f, \"%s\")\n", line, prog);
1994-05-26 05:23:31 +00:00
errno = 0; /* keep strerror() stuff out of logerror messages */
/* clear out file entry */
memset(f, 0, sizeof(*f));
for (i = 0; i <= LOG_NFACILITIES; i++)
f->f_pmask[i] = INTERNAL_NOPRI;
/* save program name if any */
if(prog && *prog=='*') prog = NULL;
if(prog) {
f->f_program = calloc(1, strlen(prog)+1);
if(f->f_program) {
strcpy(f->f_program, prog);
}
}
1994-05-26 05:23:31 +00:00
/* scan through the list of selectors */
for (p = line; *p && *p != '\t' && *p != ' ';) {
int pri_done;
int pri_cmp;
1994-05-26 05:23:31 +00:00
/* find the end of this facility name list */
for (q = p; *q && *q != '\t' && *q != ' ' && *q++ != '.'; )
1994-05-26 05:23:31 +00:00
continue;
/* get the priority comparison */
pri_cmp = 0;
pri_done = 0;
while (!pri_done) {
switch (*q) {
case '<':
pri_cmp |= PRI_LT;
q++;
break;
case '=':
pri_cmp |= PRI_EQ;
q++;
break;
case '>':
pri_cmp |= PRI_GT;
q++;
break;
default:
pri_done++;
break;
}
}
if (!pri_cmp)
pri_cmp = (UniquePriority)
? (PRI_EQ)
: (PRI_EQ | PRI_GT)
;
1994-05-26 05:23:31 +00:00
/* collect priority name */
for (bp = buf; *q && !strchr("\t,; ", *q); )
1994-05-26 05:23:31 +00:00
*bp++ = *q++;
*bp = '\0';
/* skip cruft */
while (strchr(",;", *q))
1994-05-26 05:23:31 +00:00
q++;
/* decode priority name */
if (*buf == '*')
pri = LOG_PRIMASK + 1;
else {
pri = decode(buf, prioritynames);
if (pri < 0) {
(void)snprintf(ebuf, sizeof ebuf,
1994-05-26 05:23:31 +00:00
"unknown priority name \"%s\"", buf);
logerror(ebuf);
return;
}
}
/* scan facilities */
while (*p && !strchr("\t.; ", *p)) {
for (bp = buf; *p && !strchr("\t,;. ", *p); )
1994-05-26 05:23:31 +00:00
*bp++ = *p++;
*bp = '\0';
1994-05-26 05:23:31 +00:00
if (*buf == '*')
for (i = 0; i < LOG_NFACILITIES; i++) {
1994-05-26 05:23:31 +00:00
f->f_pmask[i] = pri;
f->f_pcmp[i] = pri_cmp;
}
1994-05-26 05:23:31 +00:00
else {
i = decode(buf, facilitynames);
if (i < 0) {
(void)snprintf(ebuf, sizeof ebuf,
1994-05-26 05:23:31 +00:00
"unknown facility name \"%s\"",
buf);
logerror(ebuf);
return;
}
f->f_pmask[i >> 3] = pri;
f->f_pcmp[i >> 3] = pri_cmp;
1994-05-26 05:23:31 +00:00
}
while (*p == ',' || *p == ' ')
p++;
}
p = q;
}
/* skip to action part */
while (*p == '\t' || *p == ' ')
1994-05-26 05:23:31 +00:00
p++;
switch (*p)
{
case '@':
(void)strncpy(f->f_un.f_forw.f_hname, ++p,
sizeof(f->f_un.f_forw.f_hname)-1);
f->f_un.f_forw.f_hname[sizeof(f->f_un.f_forw.f_hname)-1] = '\0';
hp = gethostbyname(f->f_un.f_forw.f_hname);
1994-05-26 05:23:31 +00:00
if (hp == NULL) {
extern int h_errno;
logerror(hstrerror(h_errno));
break;
}
memset(&f->f_un.f_forw.f_addr, 0,
sizeof(f->f_un.f_forw.f_addr));
f->f_un.f_forw.f_addr.sin_family = AF_INET;
f->f_un.f_forw.f_addr.sin_port = LogPort;
memmove(&f->f_un.f_forw.f_addr.sin_addr, hp->h_addr, hp->h_length);
f->f_type = F_FORW;
break;
case '/':
if ((f->f_file = open(p, O_WRONLY|O_APPEND, 0)) < 0) {
f->f_type = F_UNUSED;
1994-05-26 05:23:31 +00:00
logerror(p);
break;
}
if (isatty(f->f_file)) {
if (strcmp(p, ctty) == 0)
f->f_type = F_CONSOLE;
else
f->f_type = F_TTY;
(void)strcpy(f->f_un.f_fname, p + sizeof _PATH_DEV - 1);
} else {
(void)strcpy(f->f_un.f_fname, p);
1994-05-26 05:23:31 +00:00
f->f_type = F_FILE;
}
1994-05-26 05:23:31 +00:00
break;
case '|':
f->f_un.f_pipe.f_pid = 0;
(void)strcpy(f->f_un.f_pipe.f_pname, p + 1);
f->f_type = F_PIPE;
break;
1994-05-26 05:23:31 +00:00
case '*':
f->f_type = F_WALL;
break;
default:
for (i = 0; i < MAXUNAMES && *p; i++) {
for (q = p; *q && *q != ','; )
q++;
(void)strncpy(f->f_un.f_uname[i], p, UT_NAMESIZE);
if ((q - p) > UT_NAMESIZE)
f->f_un.f_uname[i][UT_NAMESIZE] = '\0';
else
f->f_un.f_uname[i][q - p] = '\0';
while (*q == ',' || *q == ' ')
q++;
p = q;
}
f->f_type = F_USERS;
break;
}
}
/*
* Decode a symbolic name to a numeric value
*/
int
decode(name, codetab)
const char *name;
CODE *codetab;
{
CODE *c;
char *p, buf[40];
if (isdigit(*name))
return (atoi(name));
for (p = buf; *name && p < &buf[sizeof(buf) - 1]; p++, name++) {
if (isupper(*name))
*p = tolower(*name);
else
*p = *name;
}
*p = '\0';
for (c = codetab; c->c_name; c++)
if (!strcmp(buf, c->c_name))
return (c->c_val);
return (-1);
}
/*
* fork off and become a daemon, but wait for the child to come online
* before returing to the parent, or we get disk thrashing at boot etc.
* Set a timer so we don't hang forever if it wedges.
*/
int
waitdaemon(nochdir, noclose, maxwait)
int nochdir, noclose, maxwait;
{
int fd;
int status;
pid_t pid, childpid;
switch (childpid = fork()) {
case -1:
return (-1);
case 0:
break;
default:
signal(SIGALRM, timedout);
alarm(maxwait);
while ((pid = wait3(&status, 0, NULL)) != -1) {
if (WIFEXITED(status))
errx(1, "child pid %d exited with return code %d",
pid, WEXITSTATUS(status));
if (WIFSIGNALED(status))
errx(1, "child pid %d exited on signal %d%s",
pid, WTERMSIG(status),
WCOREDUMP(status) ? " (core dumped)" :
"");
if (pid == childpid) /* it's gone... */
break;
}
exit(0);
}
if (setsid() == -1)
return (-1);
if (!nochdir)
(void)chdir("/");
if (!noclose && (fd = open(_PATH_DEVNULL, O_RDWR, 0)) != -1) {
(void)dup2(fd, STDIN_FILENO);
(void)dup2(fd, STDOUT_FILENO);
(void)dup2(fd, STDERR_FILENO);
if (fd > 2)
(void)close (fd);
}
return (getppid());
}
/*
* We get a SIGALRM from the child when it's running and finished doing it's
* fsync()'s or O_SYNC writes for all the boot messages.
*
* We also get a signal from the kernel if the timer expires, so check to
* see what happened.
*/
void
timedout(sig)
int sig __unused;
{
int left;
left = alarm(0);
signal(SIGALRM, SIG_DFL);
if (left == 0)
errx(1, "timed out waiting for child");
else
exit(0);
}
/*
* Add `s' to the list of allowable peer addresses to accept messages
* from.
*
* `s' is a string in the form:
*
* [*]domainname[:{servicename|portnumber|*}]
*
* or
*
* netaddr/maskbits[:{servicename|portnumber|*}]
*
* Returns -1 on error, 0 if the argument was valid.
*/
int
allowaddr(s)
char *s;
{
char *cp1, *cp2;
struct allowedpeer ap;
struct servent *se;
regex_t re;
int i;
if ((cp1 = strrchr(s, ':'))) {
/* service/port provided */
*cp1++ = '\0';
if (strlen(cp1) == 1 && *cp1 == '*')
/* any port allowed */
ap.port = htons(0);
else if ((se = getservbyname(cp1, "udp")))
ap.port = se->s_port;
else {
ap.port = htons((int)strtol(cp1, &cp2, 0));
if (*cp2 != '\0')
return -1; /* port not numeric */
}
} else {
if ((se = getservbyname("syslog", "udp")))
ap.port = se->s_port;
else
/* sanity, should not happen */
ap.port = htons(514);
}
/* the regexp's are ugly, but the cleanest way */
if (regcomp(&re, "^[0-9]+\\.[0-9]+\\.[0-9]+\\.[0-9]+(/[0-9]+)?$",
REG_EXTENDED))
/* if RE compilation fails, that's an internal error */
abort();
if (regexec(&re, s, 0, 0, 0) == 0) {
/* arg `s' is numeric */
ap.isnumeric = 1;
if ((cp1 = strchr(s, '/')) != NULL) {
*cp1++ = '\0';
i = atoi(cp1);
if (i < 0 || i > 32)
return -1;
/* convert masklen to netmask */
ap.a_mask.s_addr = htonl(~((1 << (32 - i)) - 1));
}
if (ascii2addr(AF_INET, s, &ap.a_addr) == -1)
return -1;
if (cp1 == NULL) {
/* use default netmask */
if (IN_CLASSA(ntohl(ap.a_addr.s_addr)))
ap.a_mask.s_addr = htonl(IN_CLASSA_NET);
else if (IN_CLASSB(ntohl(ap.a_addr.s_addr)))
ap.a_mask.s_addr = htonl(IN_CLASSB_NET);
else
ap.a_mask.s_addr = htonl(IN_CLASSC_NET);
}
} else {
/* arg `s' is domain name */
ap.isnumeric = 0;
ap.a_name = s;
}
regfree(&re);
if (Debug) {
printf("allowaddr: rule %d: ", NumAllowed);
if (ap.isnumeric) {
printf("numeric, ");
printf("addr = %s, ",
addr2ascii(AF_INET, &ap.a_addr, sizeof(struct in_addr), 0));
printf("mask = %s; ",
addr2ascii(AF_INET, &ap.a_mask, sizeof(struct in_addr), 0));
} else
printf("domainname = %s; ", ap.a_name);
printf("port = %d\n", ntohs(ap.port));
}
if ((AllowedPeers = realloc(AllowedPeers,
++NumAllowed * sizeof(struct allowedpeer)))
== NULL) {
fprintf(stderr, "Out of memory!\n");
exit(EX_OSERR);
}
memcpy(&AllowedPeers[NumAllowed - 1], &ap, sizeof(struct allowedpeer));
return 0;
}
/*
* Validate that the remote peer has permission to log to us.
*/
int
validate(sin, hname)
struct sockaddr_in *sin;
const char *hname;
{
int i;
size_t l1, l2;
char *cp, name[MAXHOSTNAMELEN];
struct allowedpeer *ap;
if (NumAllowed == 0)
/* traditional behaviour, allow everything */
return 1;
strncpy(name, hname, sizeof name);
if (strchr(name, '.') == NULL) {
strncat(name, ".", sizeof name - strlen(name) - 1);
strncat(name, LocalDomain, sizeof name - strlen(name) - 1);
}
dprintf("validate: dgram from IP %s, port %d, name %s;\n",
addr2ascii(AF_INET, &sin->sin_addr, sizeof(struct in_addr), 0),
ntohs(sin->sin_port), name);
/* now, walk down the list */
for (i = 0, ap = AllowedPeers; i < NumAllowed; i++, ap++) {
if (ntohs(ap->port) != 0 && ap->port != sin->sin_port) {
dprintf("rejected in rule %d due to port mismatch.\n", i);
continue;
}
if (ap->isnumeric) {
if ((sin->sin_addr.s_addr & ap->a_mask.s_addr)
!= ap->a_addr.s_addr) {
dprintf("rejected in rule %d due to IP mismatch.\n", i);
continue;
}
} else {
cp = ap->a_name;
l1 = strlen(name);
if (*cp == '*') {
/* allow wildmatch */
cp++;
l2 = strlen(cp);
if (l2 > l1 || memcmp(cp, &name[l1 - l2], l2) != 0) {
dprintf("rejected in rule %d due to name mismatch.\n", i);
continue;
}
} else {
/* exact match */
l2 = strlen(cp);
if (l2 != l1 || memcmp(cp, name, l1) != 0) {
dprintf("rejected in rule %d due to name mismatch.\n", i);
continue;
}
}
}
dprintf("accepted in rule %d.\n", i);
return 1; /* hooray! */
}
return 0;
}
/*
* Fairly similar to popen(3), but returns an open descriptor, as
* opposed to a FILE *.
*/
int
p_open(prog, pid)
char *prog;
pid_t *pid;
{
int pfd[2], nulldesc, i;
sigset_t omask, mask;
char *argv[4]; /* sh -c cmd NULL */
char errmsg[200];
if (pipe(pfd) == -1)
return -1;
if ((nulldesc = open(_PATH_DEVNULL, O_RDWR)) == -1)
/* we are royally screwed anyway */
return -1;
sigemptyset(&mask);
sigaddset(&mask, SIGALRM);
sigaddset(&mask, SIGHUP);
sigprocmask(SIG_BLOCK, &mask, &omask);
switch ((*pid = fork())) {
case -1:
sigprocmask(SIG_SETMASK, &omask, 0);
close(nulldesc);
return -1;
case 0:
argv[0] = "sh";
argv[1] = "-c";
argv[2] = prog;
argv[3] = NULL;
alarm(0);
(void)setsid(); /* Avoid catching SIGHUPs. */
/*
* Throw away pending signals, and reset signal
* behaviour to standard values.
*/
signal(SIGALRM, SIG_IGN);
signal(SIGHUP, SIG_IGN);
sigprocmask(SIG_SETMASK, &omask, 0);
signal(SIGPIPE, SIG_DFL);
signal(SIGQUIT, SIG_DFL);
signal(SIGALRM, SIG_DFL);
signal(SIGHUP, SIG_DFL);
dup2(pfd[0], STDIN_FILENO);
dup2(nulldesc, STDOUT_FILENO);
dup2(nulldesc, STDERR_FILENO);
for (i = getdtablesize(); i > 2; i--)
(void) close(i);
(void) execvp(_PATH_BSHELL, argv);
_exit(255);
}
sigprocmask(SIG_SETMASK, &omask, 0);
close(nulldesc);
close(pfd[0]);
/*
* Avoid blocking on a hung pipe. With O_NONBLOCK, we are
* supposed to get an EWOULDBLOCK on writev(2), which is
* caught by the logic above anyway, which will in turn close
* the pipe, and fork a new logging subprocess if necessary.
* The stale subprocess will be killed some time later unless
* it terminated itself due to closing its input pipe (so we
* get rid of really dead puppies).
*/
if (fcntl(pfd[1], F_SETFL, O_NONBLOCK) == -1) {
/* This is bad. */
(void)snprintf(errmsg, sizeof errmsg,
"Warning: cannot change pipe to PID %d to "
"non-blocking behaviour.",
(int)*pid);
logerror(errmsg);
}
return pfd[1];
}
void
Fix a serious bug in syslogd regarding the handling of pipes. The bug would cause syslogd to eventually kill innocent processes in the system over time (note: not `could' but `would'). Many thanks to my colleague Mirko for digging into the kernel structures and providing me with the debugging framework to find out about the nature of this bug (and to isolate that syslogd was the culprit) in a rather large set of distributed machines at client sites where this happened occasionally. Whenever a child process was no longer responsive, or when syslogd receives a SIGHUP so it closes all its logging file descriptors, for any descriptor that refers to a pipe syslogd enters the data about the old logging child process into a `dead queue', where it is being removed from (and the status of the dead kitten being fetched) upon receipt of a SIGCHLD. However, there's a high probability that the SIGCHLD already arrives before the child's data are actually entered into the dead queue inside the SIGHUP handler, so the SIGCHLD handler has nothing to fetch and remove and simply continues. Whenever this happens, the process'es data remain on the dead queue forever, and since domark() tried to get rid of totally unresponsive children by first sending a SIGTERM and later a SIGKILL, it was only a matter of time until the system had recycled enough PIDs so an innocent process got shot to death. Fix the race by masking SIGHUP and SIGCHLD from both handlers mutually. Add additional bandaids ``just in case'', i. e. don't enter a process into the dead queue if we can't signal it (this should only happen in case it is already dead by that time so we can fetch the status immediately instead of deferring this to the SIGCHLD handler); for the kill(2) inside domark(), check for an error status (/* Can't happen */ :) and remove it from the dead queue in this case (which if it would have been there in the first place would have reduced the problem to a statistically minimal likelihood so i certainly would never have noticed the bug at all :). Mirko also reviewed the fix in priciple (mutual blocking of both signals inside the handlers), but not the actual code. Reviewed by: Mirko Kaffka <mirko@interface-business.de> Approved by: jkh
2000-02-28 17:49:43 +00:00
deadq_enter(pid, name)
pid_t pid;
Fix a serious bug in syslogd regarding the handling of pipes. The bug would cause syslogd to eventually kill innocent processes in the system over time (note: not `could' but `would'). Many thanks to my colleague Mirko for digging into the kernel structures and providing me with the debugging framework to find out about the nature of this bug (and to isolate that syslogd was the culprit) in a rather large set of distributed machines at client sites where this happened occasionally. Whenever a child process was no longer responsive, or when syslogd receives a SIGHUP so it closes all its logging file descriptors, for any descriptor that refers to a pipe syslogd enters the data about the old logging child process into a `dead queue', where it is being removed from (and the status of the dead kitten being fetched) upon receipt of a SIGCHLD. However, there's a high probability that the SIGCHLD already arrives before the child's data are actually entered into the dead queue inside the SIGHUP handler, so the SIGCHLD handler has nothing to fetch and remove and simply continues. Whenever this happens, the process'es data remain on the dead queue forever, and since domark() tried to get rid of totally unresponsive children by first sending a SIGTERM and later a SIGKILL, it was only a matter of time until the system had recycled enough PIDs so an innocent process got shot to death. Fix the race by masking SIGHUP and SIGCHLD from both handlers mutually. Add additional bandaids ``just in case'', i. e. don't enter a process into the dead queue if we can't signal it (this should only happen in case it is already dead by that time so we can fetch the status immediately instead of deferring this to the SIGCHLD handler); for the kill(2) inside domark(), check for an error status (/* Can't happen */ :) and remove it from the dead queue in this case (which if it would have been there in the first place would have reduced the problem to a statistically minimal likelihood so i certainly would never have noticed the bug at all :). Mirko also reviewed the fix in priciple (mutual blocking of both signals inside the handlers), but not the actual code. Reviewed by: Mirko Kaffka <mirko@interface-business.de> Approved by: jkh
2000-02-28 17:49:43 +00:00
const char *name;
{
dq_t p;
Fix a serious bug in syslogd regarding the handling of pipes. The bug would cause syslogd to eventually kill innocent processes in the system over time (note: not `could' but `would'). Many thanks to my colleague Mirko for digging into the kernel structures and providing me with the debugging framework to find out about the nature of this bug (and to isolate that syslogd was the culprit) in a rather large set of distributed machines at client sites where this happened occasionally. Whenever a child process was no longer responsive, or when syslogd receives a SIGHUP so it closes all its logging file descriptors, for any descriptor that refers to a pipe syslogd enters the data about the old logging child process into a `dead queue', where it is being removed from (and the status of the dead kitten being fetched) upon receipt of a SIGCHLD. However, there's a high probability that the SIGCHLD already arrives before the child's data are actually entered into the dead queue inside the SIGHUP handler, so the SIGCHLD handler has nothing to fetch and remove and simply continues. Whenever this happens, the process'es data remain on the dead queue forever, and since domark() tried to get rid of totally unresponsive children by first sending a SIGTERM and later a SIGKILL, it was only a matter of time until the system had recycled enough PIDs so an innocent process got shot to death. Fix the race by masking SIGHUP and SIGCHLD from both handlers mutually. Add additional bandaids ``just in case'', i. e. don't enter a process into the dead queue if we can't signal it (this should only happen in case it is already dead by that time so we can fetch the status immediately instead of deferring this to the SIGCHLD handler); for the kill(2) inside domark(), check for an error status (/* Can't happen */ :) and remove it from the dead queue in this case (which if it would have been there in the first place would have reduced the problem to a statistically minimal likelihood so i certainly would never have noticed the bug at all :). Mirko also reviewed the fix in priciple (mutual blocking of both signals inside the handlers), but not the actual code. Reviewed by: Mirko Kaffka <mirko@interface-business.de> Approved by: jkh
2000-02-28 17:49:43 +00:00
int status;
/*
* Be paranoid, if we can't signal the process, don't enter it
* into the dead queue (perhaps it's already dead). If possible,
* we try to fetch and log the child's status.
*/
if (kill(pid, 0) != 0) {
if (waitpid(pid, &status, WNOHANG) > 0)
log_deadchild(pid, status, name);
return;
}
p = malloc(sizeof(struct deadq_entry));
if (p == 0) {
errno = 0;
logerror("panic: out of virtual memory!");
exit(1);
}
p->dq_pid = pid;
p->dq_timeout = DQ_TIMO_INIT;
TAILQ_INSERT_TAIL(&deadq_head, p, dq_entries);
}
Fix a serious bug in syslogd regarding the handling of pipes. The bug would cause syslogd to eventually kill innocent processes in the system over time (note: not `could' but `would'). Many thanks to my colleague Mirko for digging into the kernel structures and providing me with the debugging framework to find out about the nature of this bug (and to isolate that syslogd was the culprit) in a rather large set of distributed machines at client sites where this happened occasionally. Whenever a child process was no longer responsive, or when syslogd receives a SIGHUP so it closes all its logging file descriptors, for any descriptor that refers to a pipe syslogd enters the data about the old logging child process into a `dead queue', where it is being removed from (and the status of the dead kitten being fetched) upon receipt of a SIGCHLD. However, there's a high probability that the SIGCHLD already arrives before the child's data are actually entered into the dead queue inside the SIGHUP handler, so the SIGCHLD handler has nothing to fetch and remove and simply continues. Whenever this happens, the process'es data remain on the dead queue forever, and since domark() tried to get rid of totally unresponsive children by first sending a SIGTERM and later a SIGKILL, it was only a matter of time until the system had recycled enough PIDs so an innocent process got shot to death. Fix the race by masking SIGHUP and SIGCHLD from both handlers mutually. Add additional bandaids ``just in case'', i. e. don't enter a process into the dead queue if we can't signal it (this should only happen in case it is already dead by that time so we can fetch the status immediately instead of deferring this to the SIGCHLD handler); for the kill(2) inside domark(), check for an error status (/* Can't happen */ :) and remove it from the dead queue in this case (which if it would have been there in the first place would have reduced the problem to a statistically minimal likelihood so i certainly would never have noticed the bug at all :). Mirko also reviewed the fix in priciple (mutual blocking of both signals inside the handlers), but not the actual code. Reviewed by: Mirko Kaffka <mirko@interface-business.de> Approved by: jkh
2000-02-28 17:49:43 +00:00
int
deadq_remove(pid)
pid_t pid;
{
dq_t q;
for (q = TAILQ_FIRST(&deadq_head); q != NULL; q = TAILQ_NEXT(q, dq_entries))
if (q->dq_pid == pid) {
TAILQ_REMOVE(&deadq_head, q, dq_entries);
free(q);
return 1;
}
return 0;
}
void
log_deadchild(pid, status, name)
pid_t pid;
int status;
const char *name;
{
int code;
char buf[256];
const char *reason;
errno = 0; /* Keep strerror() stuff out of logerror messages. */
if (WIFSIGNALED(status)) {
reason = "due to signal";
code = WTERMSIG(status);
} else {
reason = "with status";
code = WEXITSTATUS(status);
if (code == 0)
return;
}
(void)snprintf(buf, sizeof buf,
"Logging subprocess %d (%s) exited %s %d.",
pid, name, reason, code);
logerror(buf);
}