SYNOPSIS
--------
-*accesslog* [*-hFa*] [*-f* 'FORMAT'] [*-p* 'PIDFILE'] 'OUTFILE' 'CHILD' ['ARGS'...]
+*accesslog* [*-hFaL*] [*-f* 'FORMAT'] [*-p* 'PIDFILE'] 'OUTFILE' 'CHILD' ['ARGS'...]
+*accesslog* *-P* 'LOGFILE'
DESCRIPTION
-----------
If the child handler exits, *accesslog* exits as well.
+Normally, *accesslog* locks the logfile using *fcntl*(2) to ensure
+that only one process writes to a logfile at any time. The *-L* switch
+can be used to override that behavior to let several processes share a
+logfile, or to use logfiles that cannot be locked for some reason.
+
OPTIONS
-------
%A - - [%{%d/%b/%Y:%H:%M:%S %z}t] "%m %u %v" - - "%R" "%G"
--------
+*-L*::
+
+ Do not attempt to lock the logfile. Note that this switch
+ conflicts with the use of the *-P* option.
+
+*-P* 'LOGFILE'::
+
+ Makes *accesslog* fetch the PID of the process currently
+ holding the lock on 'LOGFILE', write that to standard output,
+ and then exit. No further command-line arguments are
+ processed. This option is useful for sending SIGHUP to
+ accesslog when rotating logfiles without having to use a PID
+ file.
+
FORMAT
------
#include <sys/time.h>
#include <signal.h>
#include <fcntl.h>
+#include <sys/stat.h>
#ifdef HAVE_CONFIG_H
#include <config.h>
static int ch;
static char *outname = NULL;
static FILE *out;
-static int flush = 1;
+static int flush = 1, locklog = 1;
static char *format;
static struct timeval now;
static volatile int reopen = 0;
reopen = 1;
}
+static int lockfile(FILE *file)
+{
+ struct flock ld;
+
+ memset(&ld, 0, sizeof(ld));
+ ld.l_type = F_WRLCK;
+ ld.l_whence = SEEK_SET;
+ ld.l_start = 0;
+ ld.l_len = 0;
+ return(fcntl(fileno(file), F_SETLK, &ld));
+}
+
+static void fetchpid(char *filename)
+{
+ int fd, ret;
+ struct flock ld;
+
+ if((fd = open(filename, O_WRONLY)) < 0) {
+ fprintf(stderr, "accesslog: %s: %s\n", filename, strerror(errno));
+ exit(1);
+ }
+ memset(&ld, 0, sizeof(ld));
+ ld.l_type = F_WRLCK;
+ ld.l_whence = SEEK_SET;
+ ld.l_start = 0;
+ ld.l_len = 0;
+ ret = fcntl(fd, F_GETLK, &ld);
+ close(fd);
+ if(ret) {
+ fprintf(stderr, "accesslog: %s: %s\n", filename, strerror(errno));
+ exit(1);
+ }
+ if(ld.l_type == F_UNLCK) {
+ fprintf(stderr, "accesslog: %s: not locked\n", filename);
+ exit(1);
+ }
+ printf("%i\n", (int)ld.l_pid);
+}
+
static void reopenlog(void)
{
FILE *new;
+ struct stat olds, news;
if(outname == NULL) {
flog(LOG_WARNING, "accesslog: received SIGHUP but logging to stdout, so ignoring");
return;
}
+ if(locklog) {
+ if(fstat(fileno(out), &olds)) {
+ flog(LOG_ERR, "accesslog: could not stat current logfile(?!): %s", strerror(errno));
+ return;
+ }
+ if(!stat(outname, &news)) {
+ if((olds.st_dev == news.st_dev) && (olds.st_ino == news.st_ino)) {
+ /*
+ * This needs to be ignored, because if the same logfile
+ * is opened and then closed, the lock is lost. To quote
+ * the Linux fcntl(2) manpage: "This is bad." No kidding.
+ *
+ * Technically, there is a race condition here when the
+ * file has been stat'ed but not yet opened, where the old
+ * log file, having been previously renamed, changes name
+ * back to the name accesslog knows and is thus reopened
+ * regardlessly, but I think that might fit under the
+ * idiom "pathological case". It should, at least, not be
+ * a security problem.
+ */
+ flog(LOG_INFO, "accesslog: received SIGHUP, but logfile has not changed, so ignoring");
+ return;
+ }
+ }
+ }
if((new = fopen(outname, "a")) == NULL) {
flog(LOG_WARNING, "accesslog: could not reopen log file `%s' on SIGHUP: %s", outname, strerror(errno));
return;
}
+ if(locklog) {
+ if(lockfile(new)) {
+ if((errno == EAGAIN) || (errno == EACCES)) {
+ flog(LOG_ERR, "accesslog: logfile is already locked; reverting to current log", strerror(errno));
+ fclose(new);
+ return;
+ } else {
+ flog(LOG_WARNING, "accesslog: could not lock logfile, so no lock will be held: %s", strerror(errno));
+ }
+ }
+ }
fclose(out);
out = new;
}
static void usage(FILE *out)
{
- fprintf(out, "usage: accesslog [-hFa] [-f FORMAT] [-p PIDFILE] OUTFILE CHILD [ARGS...]\n");
+ fprintf(out, "usage: accesslog [-hFaL] [-f FORMAT] [-p PIDFILE] OUTFILE CHILD [ARGS...]\n");
}
int main(int argc, char **argv)
FILE *pidout;
pidfile = NULL;
- while((c = getopt(argc, argv, "+hFaf:p:")) >= 0) {
+ while((c = getopt(argc, argv, "+hFaLf:p:P:")) >= 0) {
switch(c) {
case 'h':
usage(stdout);
case 'F':
flush = 0;
break;
+ case 'L':
+ locklog = 0;
+ break;
case 'f':
format = optarg;
break;
+ case 'P':
+ fetchpid(optarg);
+ exit(0);
case 'p':
pidfile = optarg;
break;
exit(1);
}
}
+ if(locklog) {
+ if(lockfile(out)) {
+ if((errno == EAGAIN) || (errno == EACCES)) {
+ flog(LOG_ERR, "accesslog: logfile is already locked", strerror(errno));
+ exit(1);
+ } else {
+ flog(LOG_WARNING, "accesslog: could not lock logfile: %s", strerror(errno));
+ }
+ }
+ }
if((ch = stdmkchild(argv + optind + 1, NULL, NULL)) < 0) {
flog(LOG_ERR, "accesslog: could not fork child: %s", strerror(errno));
exit(1);