libslack(daemon) - daemon module
#include <slack/daemon.h>
typedef void daemon_config_parser_t(void *obj, const char *path, char *line, size_t lineno);
int daemon_started_by_init(void); int daemon_started_by_inetd(void); int daemon_prevent_core(void); int daemon_revoke_privileges(void); int daemon_become_user(uid_t uid, gid_t gid, char *user); char *daemon_absolute_path(const char *path); int daemon_path_is_safe(const char *path); void *daemon_parse_config(const char *path, void *obj, daemon_config_parser_t *parser); int daemon_init(const char *name); int daemon_close(void);
This module provides functions for writing daemons. There are many tasks that need to be performed to correctly set up a daemon process. This can be tedious. These functions perform these tasks for you.
int daemon_started_by_init(void)
If this process was started by init(8), returns 1. If not, returns 0
. If it was, we might be getting respawned so fork(2) and exit(2) would be a big mistake (and unnecessary anyway since there is no
controlling terminal). On error, returns -1
with errno
set appropriately.
int daemon_started_by_inetd(void)
If this process was started by inetd(8), returns 1
. If not, returns
0
. On error, returns -1
with errno
set appropriately. If it was,
stdin
, stdout
and stderr
would be opened to a socket. Closing them would be a big mistake. We also
would not need to fork(2) and exit(2)
because there is no controlling terminal.
int daemon_prevent_core(void)
Prevents core files from being generated. This is used to prevent leaking
sensitive information in daemons run by root. On success, returns 0
. On error, returns -1
with errno
set appropriately.
int daemon_revoke_privileges(void)
Revokes setuid and setgid privileges. Useful when your program does not
require any special privileges and may become unsafe if incorrectly
installed with special privileges. Also useful when your program only
requires special privileges upon startup (e.g. binding to a privileged
socket). Performs the following: Sets the effective gid to the real gid if
they differ. Checks that they no longer differ. Sets the effective uid to
the real uid if they differ. Checks that they no longer differ. Also closes
/etc/passwd and /etc/group in case they were opened by root and give access
to user and group passwords. On success, returns 0
. On error, returns
-1
with errno
set appropriately.
int daemon_become_user(uid_t uid, gid_t gid, char *user)
Changes the owner and group of the process to uid
and gid
respectively. If user
is not null, the supplementary group list will be initialised with initgroups(). Otherwise, the supplementary group list will be cleared of all groups. On
success, returns 0. On error, returns -1. Only root can use this function.
char *daemon_absolute_path(const char *path)
Returns path
converted into an absolute path. Cleans up any .
and
..
and //
and trailing /
found in the returned path. Note that the returned path looks canonical but
isn't because symbolic links are not followed and expanded. It is the
caller's responsibility to deallocate the path returned with mem_release(3) or free(3). On success, returns the absolute path. On error, returns null
with errno
set appropriately.
int daemon_path_is_safe(const char *path)
Checks that the file referred to by path
is not group or world writable. Also checks that the containing directories
are not group or world writable, following symbolic links. Useful when you
need to know whether or not you can trust a user supplied
configuration/command file before reading and acting upon its contents. On
success, returns 1 if path
is safe or 0 if it is not. On error, returns -1
with errno
set appropriately.
void *daemon_parse_config(const char *path, void *obj, daemon_config_parser_t *parser)
Parses the text configuration file named path
. Blank lines are ignored. Comments ('#'
to end of line) are ignored. Lines that end with '\'
are joined with the following line. There may be whitespace and even a
comment after the '\'
character but nothing else. The parser
function is called with the client supplied obj
, the file name, the line and the line number as arguments. On success,
returns obj
. On error, returns null
(i.e. if the configuration file could not be read). Note: Don't parse
config files unless they are ``safe'' as determined by daemon_path_is_safe().
int daemon_init(const char *name)
Initialises a daemon by performing the following tasks:
If the process was not invoked by init(8) or inetd(8):
Background the process to lose process group leadership.
Start a new process session.
Under SVR4, background the process again to lose process session leadership. This
prevents the process from ever gaining a controlling terminal. This only
happens when SVR4
is defined and
NO_EXTRA_SVR4_FORK
is not defined when libslack is compiled. Before doing this, ignore SIGHUP
because when the session leader terminates, all processes in the foreground
process group are sent a SIGHUP
signal (apparently). Note that this code may not execute (e.g. when started
by
init(8) or inetd(8) or when either SVR4
was not defined or
NO_EXTRA_SVR4_FORK
was defined when libslack was compiled). This means that the client can't make any assumptions about
the SIGHUP
handler when
daemon_init() returns.
Change directory to the root directory so as not to hamper umounts.
Clear the umask to enable explicit file creation modes.
Close all open file descriptors. If the process was invoked by inetd(8),
stdin
, stdout
and stderr
are left open since they are open to a socket.
Open stdin
, stdout
and stderr
to /dev/null
in case something requires them to be open. Of course, this is not done if
the process was invoked by inetd(8).
If name
is non-null, create and lock a file containing the process id of the
process. The presence of this locked file prevents two instances of a
daemon with the same name from running at the same time. The default
location of the pidfile is /var/run
on Linux and /etc
on Solaris for
root or /tmp
for ordinary users.
On success, returns 0
. On error, returns -1
with errno
set appropriately.
int daemon_close(void)
Unlinks the locked pid file, if any. Returns 0.
Additional errors may be generated and returned from the underlying system calls. See their manual pages.
An argument was invalid (e.g. null
).
The name
passed to daemon_init() or daemon_path_is_safe() resulted in a path name that is too long for the intended filesystem.
daemon_path_is_safe() recursed too deeply (16 levels).
#include <stdio.h> #include <unistd.h> #include <syslog.h>
#include <slack/lib.h>
const char * const config_fname = "/etc/fstab"; List *config = NULL;
void fstab_parser(void *obj, const char *path, char *line, size_t lineno) { char device[64], mount[64], fstype[64], opts[64]; int freq, passno;
if (sscanf(line, "%s %s %s %s %d %d", device, mount, fstype, opts, &freq, &passno) != 6) fprintf(stderr, "Syntax error in %s (line %d): %s\n", path, lineno, line); else { char *copy;
printf("%s %s %s %s %d %d\n", device, mount, fstype, opts, freq, passno);
if (!(copy = mem_strdup(line))) fprintf(stderr, "out of memory\n"); else if (!list_append(config, copy)) fprintf(stderr, "failed to add line %d to config\n", lineno); } }
void hup(int signo) { list_remove_range(config, 0, -1); daemon_parse_config(config_fname, config, fstab_parser); }
void term(int signo) { daemon_close(); exit(EXIT_SUCCESS); }
void do_stuff() { // do stuff... kill(getpid(), SIGTERM); signal_handle_all(); }
int main(int ac, char **av) { if (daemon_revoke_privileges() == -1 || daemon_prevent_core() == -1 || daemon_path_is_safe(config_fname) != 1 || (config = list_create(free)) == NULL || daemon_parse_config(config_fname, config, fstab_parser) == NULL || daemon_init(prog_basename(*av)) == -1 || signal_set_handler(SIGHUP, 0, hup) == -1 || signal_set_handler(SIGTERM, 0, term) == -1) return 1;
do_stuff();
return EXIT_SUCCESS; // unreached }
MT-Safe
It is possible to obtain a controlling terminal under BSD (and even under
SVR4 if SVR4 was not defined or NO_EXTRA_SVR4_FORK
was defined when
libslack is compiled). If anything calls open(2) on a terminal device without the O_NOCTTY
flag, the process doing so will obtain a controlling terminal.
Because root's pidfiles are created in a different directory (/var/run
on Linux, /etc
on Solaris) to those of ordinary users (/tmp
), it is possible for root and another user to use the same name for a daemon client. This shouldn't
be a problem but if it is, recompile libslack and relink daemon so that all pidfiles are created in /tmp
by defining
ROOT_PID_DIR
and USER_PID_DIR
to both be /tmp
.
The exclusive creation and locking of the pidfile doesn't work correctly over NFS on Linux so pidfiles must reside locally.
daemon_path_is_safe() ignores ACLs (so does sendmail(8)). It should probably treat a path as unsafe if there are any ACLs (allowing extra access) along the path.
The functions daemon_prevent_core(), daemon_revoke_privileges(), daemon_become_user(), daemon_absolute_path(), daemon_path_is_safe() and daemon_parse_config() should probably all have the daemon_ prefix removed from their names. Their use is more general than just in daemons.
libslack(3), daemon(1), init(8), inetd(8), fork(2), umask(2), setsid(2), chdir(2), setrlimit(2), setgid(2), setuid(2), setgroups(2), initgroups(3), endpwent(3), endgrent(3)
20011109 raf <raf@raf.org>