/* Inner loops of cache daemon.
- Copyright (C) 1998-2007, 2008 Free Software Foundation, Inc.
+ Copyright (C) 1998-2016 Free Software Foundation, Inc.
This file is part of the GNU C Library.
Contributed by Ulrich Drepper <drepper@cygnus.com>, 1998.
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
- along with this program; if not, write to the Free Software Foundation,
- Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */
+ along with this program; if not, see <http://www.gnu.org/licenses/>. */
#include <alloca.h>
#include <assert.h>
#include <errno.h>
#include <fcntl.h>
#include <grp.h>
+#include <ifaddrs.h>
#include <libintl.h>
#include <pthread.h>
#include <pwd.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
+#include <stdint.h>
#include <arpa/inet.h>
+#ifdef HAVE_NETLINK
+# include <linux/netlink.h>
+# include <linux/rtnetlink.h>
+#endif
#ifdef HAVE_EPOLL
# include <sys/epoll.h>
#endif
#include "dbg_log.h"
#include "selinux.h"
#include <resolv/resolv.h>
-#ifdef HAVE_SENDFILE
-# include <kernel-features.h>
-#endif
+#include <kernel-features.h>
+#include <libc-internal.h>
-/* Wrapper functions with error checking for standard functions. */
-extern void *xmalloc (size_t n);
-extern void *xcalloc (size_t n, size_t s);
-extern void *xrealloc (void *o, size_t n);
/* Support to run nscd as an unprivileged user */
const char *server_user;
[INITGROUPS] = "INITGROUPS",
[GETSERVBYNAME] = "GETSERVBYNAME",
[GETSERVBYPORT] = "GETSERVBYPORT",
- [GETFDSERV] = "GETFDSERV"
+ [GETFDSERV] = "GETFDSERV",
+ [GETNETGRENT] = "GETNETGRENT",
+ [INNETGR] = "INNETGR",
+ [GETFDNETGR] = "GETFDNETGR"
};
/* The control data structures for the services. */
[pwddb] = {
.lock = PTHREAD_RWLOCK_WRITER_NONRECURSIVE_INITIALIZER_NP,
.prune_lock = PTHREAD_MUTEX_INITIALIZER,
+ .prune_run_lock = PTHREAD_MUTEX_INITIALIZER,
.enabled = 0,
.check_file = 1,
.persistent = 0,
.shared = 0,
.max_db_size = DEFAULT_MAX_DB_SIZE,
.suggested_module = DEFAULT_SUGGESTED_MODULE,
- .reset_res = 0,
- .filename = "/etc/passwd",
.db_filename = _PATH_NSCD_PASSWD_DB,
.disabled_iov = &pwd_iov_disabled,
.postimeout = 3600,
[grpdb] = {
.lock = PTHREAD_RWLOCK_WRITER_NONRECURSIVE_INITIALIZER_NP,
.prune_lock = PTHREAD_MUTEX_INITIALIZER,
+ .prune_run_lock = PTHREAD_MUTEX_INITIALIZER,
.enabled = 0,
.check_file = 1,
.persistent = 0,
.shared = 0,
.max_db_size = DEFAULT_MAX_DB_SIZE,
.suggested_module = DEFAULT_SUGGESTED_MODULE,
- .reset_res = 0,
- .filename = "/etc/group",
.db_filename = _PATH_NSCD_GROUP_DB,
.disabled_iov = &grp_iov_disabled,
.postimeout = 3600,
[hstdb] = {
.lock = PTHREAD_RWLOCK_WRITER_NONRECURSIVE_INITIALIZER_NP,
.prune_lock = PTHREAD_MUTEX_INITIALIZER,
+ .prune_run_lock = PTHREAD_MUTEX_INITIALIZER,
.enabled = 0,
.check_file = 1,
.persistent = 0,
.shared = 0,
.max_db_size = DEFAULT_MAX_DB_SIZE,
.suggested_module = DEFAULT_SUGGESTED_MODULE,
- .reset_res = 1,
- .filename = "/etc/hosts",
.db_filename = _PATH_NSCD_HOSTS_DB,
.disabled_iov = &hst_iov_disabled,
.postimeout = 3600,
[servdb] = {
.lock = PTHREAD_RWLOCK_WRITER_NONRECURSIVE_INITIALIZER_NP,
.prune_lock = PTHREAD_MUTEX_INITIALIZER,
+ .prune_run_lock = PTHREAD_MUTEX_INITIALIZER,
.enabled = 0,
.check_file = 1,
.persistent = 0,
.shared = 0,
.max_db_size = DEFAULT_MAX_DB_SIZE,
.suggested_module = DEFAULT_SUGGESTED_MODULE,
- .reset_res = 0,
- .filename = "/etc/services",
.db_filename = _PATH_NSCD_SERVICES_DB,
.disabled_iov = &serv_iov_disabled,
.postimeout = 28800,
.wr_fd = -1,
.ro_fd = -1,
.mmap_used = false
+ },
+ [netgrdb] = {
+ .lock = PTHREAD_RWLOCK_WRITER_NONRECURSIVE_INITIALIZER_NP,
+ .prune_lock = PTHREAD_MUTEX_INITIALIZER,
+ .prune_run_lock = PTHREAD_MUTEX_INITIALIZER,
+ .enabled = 0,
+ .check_file = 1,
+ .persistent = 0,
+ .propagate = 0, /* Not used. */
+ .shared = 0,
+ .max_db_size = DEFAULT_MAX_DB_SIZE,
+ .suggested_module = DEFAULT_SUGGESTED_MODULE,
+ .db_filename = _PATH_NSCD_NETGROUP_DB,
+ .disabled_iov = &netgroup_iov_disabled,
+ .postimeout = 28800,
+ .negtimeout = 20,
+ .wr_fd = -1,
+ .ro_fd = -1,
+ .mmap_used = false
}
};
[INITGROUPS] = { true, &dbs[grpdb] },
[GETSERVBYNAME] = { true, &dbs[servdb] },
[GETSERVBYPORT] = { true, &dbs[servdb] },
- [GETFDSERV] = { false, &dbs[servdb] }
+ [GETFDSERV] = { false, &dbs[servdb] },
+ [GETNETGRENT] = { true, &dbs[netgrdb] },
+ [INNETGR] = { true, &dbs[netgrdb] },
+ [GETFDNETGR] = { false, &dbs[netgrdb] }
};
#ifdef HAVE_INOTIFY
/* Inotify descriptor. */
-static int inotify_fd = -1;
+int inotify_fd = -1;
+#endif
-/* Watch descriptor for resolver configuration file. */
-static int resolv_conf_descr = -1;
+#ifdef HAVE_NETLINK
+/* Descriptor for netlink status updates. */
+static int nl_status_fd = -1;
#endif
-#ifndef __ASSUME_SOCK_CLOEXEC
-/* Negative if SOCK_CLOEXEC is not supported, positive if it is, zero
- before be know the result. */
-static int have_sock_cloexec;
-/* The paccept syscall was introduced at the same time as SOCK_CLOEXEC. */
-# define have_paccept have_sock_cloexec
+#ifndef __ASSUME_ACCEPT4
+static int have_accept4;
#endif
/* Number of times clients had to wait. */
unsigned long int client_queued;
-/* Data structure for recording in-flight memory allocation. */
-__thread struct mem_in_flight mem_in_flight attribute_tls_model_ie;
-/* Global list of the mem_in_flight variables of all the threads. */
-struct mem_in_flight *mem_in_flight_list;
-
ssize_t
writeall (int fd, const void *buf, size_t len)
use_he = 1,
use_he_begin = use_he | use_begin,
use_he_end = use_he | use_end,
-#if SEPARATE_KEY
- use_key = 2,
- use_key_begin = use_key | use_begin,
- use_key_end = use_key | use_end,
- use_key_first = use_key_begin | use_first,
-#endif
use_data = 3,
use_data_begin = use_data | use_begin,
use_data_end = use_data | use_end,
static int
verify_persistent_db (void *mem, struct database_pers_head *readhead, int dbnr)
{
- assert (dbnr == pwddb || dbnr == grpdb || dbnr == hstdb || dbnr == servdb);
+ assert (dbnr == pwddb || dbnr == grpdb || dbnr == hstdb || dbnr == servdb
+ || dbnr == netgrdb);
time_t now = time (NULL);
if (here->key < here->packet + sizeof (struct datahead)
|| here->key > here->packet + dh->allocsize
|| here->key + here->len > here->packet + dh->allocsize)
- {
-#if SEPARATE_KEY
- /* If keys can appear outside of data, this should be done
- instead. But gc doesn't mark the data in that case. */
- if (! check_use (data, head->first_free, usemap,
- use_key | (here->first ? use_first : 0),
- here->key, here->len))
-#endif
- goto fail;
- }
+ goto fail;
work = here->next;
he->first == true hashentry. */
for (ref_t idx = 0; idx < head->first_free; ++idx)
{
-#if SEPARATE_KEY
- if (usemap[idx] == use_key_begin)
- goto fail;
-#endif
if (usemap[idx] == use_data_begin)
goto fail;
}
/* No configuration for this value, assume a default. */
nthreads = 4;
-#ifdef HAVE_INOTIFY
- /* Use inotify to recognize changed files. */
- inotify_fd = inotify_init1 (IN_NONBLOCK);
-# ifndef __ASSUME_IN_NONBLOCK
- if (inotify_fd == -1 && errno == ENOSYS)
- {
- inotify_fd = inotify_init ();
- if (inotify_fd != -1)
- fcntl (inotify_fd, F_SETFL, O_NONBLOCK);
- }
-# endif
-#endif
-
for (size_t cnt = 0; cnt < lastdb; ++cnt)
if (dbs[cnt].enabled)
{
if (fd != -1)
close (fd);
}
+ else if (errno == EACCES)
+ do_exit (EXIT_FAILURE, 0, _("cannot access '%s'"),
+ dbs[cnt].db_filename);
}
if (dbs[cnt].head == NULL)
{
dbg_log (_("database for %s corrupted or simultaneously used; remove %s manually if necessary and restart"),
dbnames[cnt], dbs[cnt].db_filename);
- // XXX Correct way to terminate?
- exit (1);
+ do_exit (1, 0, NULL);
}
if (dbs[cnt].persistent)
cannot create read-only descriptor for \"%s\"; no mmap"),
dbs[cnt].db_filename);
- /* Before we create the header, initialiye the hash
- table. So that if we get interrupted if writing
+ /* Before we create the header, initialize the hash
+ table. That way if we get interrupted while writing
the header we can recognize a partially initialized
database. */
size_t ps = sysconf (_SC_PAGESIZE);
dbs[cnt].shared = 0;
assert (dbs[cnt].ro_fd == -1);
}
-
- dbs[cnt].inotify_descr = -1;
- if (dbs[cnt].check_file)
- {
-#ifdef HAVE_INOTIFY
- if (inotify_fd < 0
- || (dbs[cnt].inotify_descr
- = inotify_add_watch (inotify_fd, dbs[cnt].filename,
- IN_DELETE_SELF | IN_MODIFY)) < 0)
- /* We cannot notice changes in the main thread. */
-#endif
- {
- /* We need the modification date of the file. */
- struct stat64 st;
-
- if (stat64 (dbs[cnt].filename, &st) < 0)
- {
- /* We cannot stat() the file, disable file checking. */
- dbg_log (_("cannot stat() file `%s': %s"),
- dbs[cnt].filename, strerror (errno));
- dbs[cnt].check_file = 0;
- }
- else
- dbs[cnt].file_mtime = st.st_mtime;
- }
- }
-
-#ifdef HAVE_INOTIFY
- if (cnt == hstdb && inotify_fd >= -1)
- /* We also monitor the resolver configuration file. */
- resolv_conf_descr = inotify_add_watch (inotify_fd,
- _PATH_RESCONF,
- IN_DELETE_SELF | IN_MODIFY);
-#endif
}
/* Create the socket. */
-#ifndef __ASSUME_SOCK_CLOEXEC
- sock = -1;
- if (have_sock_cloexec >= 0)
-#endif
- {
- sock = socket (AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC | SOCK_NONBLOCK, 0);
-#ifdef __ASSUME_SOCK_CLOEXEC
- if (have_sock_cloexec == 0)
- have_sock_cloexec = sock != -1 || errno != EINVAL ? 1 : -1;
-#endif
- }
-#ifndef __ASSUME_SOCK_CLOEXEC
- if (have_sock_cloexec < 0)
- sock = socket (AF_UNIX, SOCK_STREAM, 0);
-#endif
+ sock = socket (AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC | SOCK_NONBLOCK, 0);
if (sock < 0)
{
dbg_log (_("cannot open socket: %s"), strerror (errno));
- exit (errno == EACCES ? 4 : 1);
+ do_exit (errno == EACCES ? 4 : 1, 0, NULL);
}
/* Bind a name to the socket. */
struct sockaddr_un sock_addr;
if (bind (sock, (struct sockaddr *) &sock_addr, sizeof (sock_addr)) < 0)
{
dbg_log ("%s: %s", _PATH_NSCDSOCKET, strerror (errno));
- exit (errno == EACCES ? 4 : 1);
+ do_exit (errno == EACCES ? 4 : 1, 0, NULL);
}
-#ifndef __ASSUME_SOCK_CLOEXEC
- if (have_sock_cloexec < 0)
- {
- /* We don't want to get stuck on accept. */
- int fl = fcntl (sock, F_GETFL);
- if (fl == -1 || fcntl (sock, F_SETFL, fl | O_NONBLOCK) == -1)
- {
- dbg_log (_("cannot change socket to nonblocking mode: %s"),
- strerror (errno));
- exit (1);
- }
-
- /* The descriptor needs to be closed on exec. */
- if (paranoia && fcntl (sock, F_SETFD, FD_CLOEXEC) == -1)
- {
- dbg_log (_("cannot set socket to close on exec: %s"),
- strerror (errno));
- exit (1);
- }
- }
-#endif
-
/* Set permissions for the socket. */
chmod (_PATH_NSCDSOCKET, DEFFILEMODE);
{
dbg_log (_("cannot enable socket to accept connections: %s"),
strerror (errno));
- exit (1);
+ do_exit (1, 0, NULL);
+ }
+
+#ifdef HAVE_NETLINK
+ if (dbs[hstdb].enabled)
+ {
+ /* Try to open netlink socket to monitor network setting changes. */
+ nl_status_fd = socket (AF_NETLINK,
+ SOCK_RAW | SOCK_CLOEXEC | SOCK_NONBLOCK,
+ NETLINK_ROUTE);
+ if (nl_status_fd != -1)
+ {
+ struct sockaddr_nl snl;
+ memset (&snl, '\0', sizeof (snl));
+ snl.nl_family = AF_NETLINK;
+ /* XXX Is this the best set to use? */
+ snl.nl_groups = (RTMGRP_IPV4_IFADDR | RTMGRP_TC | RTMGRP_IPV4_MROUTE
+ | RTMGRP_IPV4_ROUTE | RTMGRP_IPV4_RULE
+ | RTMGRP_IPV6_IFADDR | RTMGRP_IPV6_MROUTE
+ | RTMGRP_IPV6_ROUTE | RTMGRP_IPV6_IFINFO
+ | RTMGRP_IPV6_PREFIX);
+
+ if (bind (nl_status_fd, (struct sockaddr *) &snl, sizeof (snl)) != 0)
+ {
+ close (nl_status_fd);
+ nl_status_fd = -1;
+ }
+ else
+ {
+ /* Start the timestamp process. */
+ dbs[hstdb].head->extra_data[NSCD_HST_IDX_CONF_TIMESTAMP]
+ = __bump_nl_timestamp ();
+ }
+ }
}
+#endif
- /* Change to unprivileged uid/gid/groups if specifed in config file */
+ /* Change to unprivileged uid/gid/groups if specified in config file */
if (server_user != NULL)
finish_drop_privileges ();
}
+#ifdef HAVE_INOTIFY
+#define TRACED_FILE_MASK (IN_DELETE_SELF | IN_CLOSE_WRITE | IN_MOVE_SELF)
+#define TRACED_DIR_MASK (IN_DELETE_SELF | IN_CREATE | IN_MOVED_TO | IN_MOVE_SELF)
+void
+install_watches (struct traced_file *finfo)
+{
+ /* Use inotify support if we have it. */
+ if (finfo->inotify_descr[TRACED_FILE] < 0)
+ finfo->inotify_descr[TRACED_FILE] = inotify_add_watch (inotify_fd,
+ finfo->fname,
+ TRACED_FILE_MASK);
+ if (finfo->inotify_descr[TRACED_FILE] < 0)
+ {
+ dbg_log (_("disabled inotify-based monitoring for file `%s': %s"),
+ finfo->fname, strerror (errno));
+ return;
+ }
+ dbg_log (_("monitoring file `%s` (%d)"),
+ finfo->fname, finfo->inotify_descr[TRACED_FILE]);
+ /* Additionally listen for events in the file's parent directory.
+ We do this because the file to be watched might be
+ deleted and then added back again. When it is added back again
+ we must re-add the watch. We must also cover IN_MOVED_TO to
+ detect a file being moved into the directory. */
+ if (finfo->inotify_descr[TRACED_DIR] < 0)
+ finfo->inotify_descr[TRACED_DIR] = inotify_add_watch (inotify_fd,
+ finfo->dname,
+ TRACED_DIR_MASK);
+ if (finfo->inotify_descr[TRACED_DIR] < 0)
+ {
+ dbg_log (_("disabled inotify-based monitoring for directory `%s': %s"),
+ finfo->fname, strerror (errno));
+ return;
+ }
+ dbg_log (_("monitoring directory `%s` (%d)"),
+ finfo->dname, finfo->inotify_descr[TRACED_DIR]);
+}
+#endif
+
+/* Register the file in FINFO as a traced file for the database DBS[DBIX].
+
+ We support registering multiple files per database. Each call to
+ register_traced_file adds to the list of registered files.
+
+ When we prune the database, either through timeout or a request to
+ invalidate, we will check to see if any of the registered files has changed.
+ When we accept new connections to handle a cache request we will also
+ check to see if any of the registered files has changed.
+
+ If we have inotify support then we install an inotify fd to notify us of
+ file deletion or modification, both of which will require we invalidate
+ the cache for the database. Without inotify support we stat the file and
+ store st_mtime to determine if the file has been modified. */
+void
+register_traced_file (size_t dbidx, struct traced_file *finfo)
+{
+ /* If the database is disabled or file checking is disabled
+ then ignore the registration. */
+ if (! dbs[dbidx].enabled || ! dbs[dbidx].check_file)
+ return;
+
+ if (__glibc_unlikely (debug_level > 0))
+ dbg_log (_("monitoring file %s for database %s"),
+ finfo->fname, dbnames[dbidx]);
+
+#ifdef HAVE_INOTIFY
+ install_watches (finfo);
+#endif
+ struct stat64 st;
+ if (stat64 (finfo->fname, &st) < 0)
+ {
+ /* We cannot stat() the file. Set mtime to zero and try again later. */
+ dbg_log (_("stat failed for file `%s'; will try again later: %s"),
+ finfo->fname, strerror (errno));
+ finfo->mtime = 0;
+ }
+ else
+ finfo->mtime = st.st_mtime;
+
+ /* Queue up the file name. */
+ finfo->next = dbs[dbidx].traced_files;
+ dbs[dbidx].traced_files = finfo;
+}
+
/* Close the connections. */
void
for (number = pwddb; number < lastdb; ++number)
if (strcmp (key, dbnames[number]) == 0)
{
- if (dbs[number].reset_res)
- res_init ();
-
+ struct traced_file *runp = dbs[number].traced_files;
+ while (runp != NULL)
+ {
+ /* Make sure we reload from file when checking mtime. */
+ runp->mtime = 0;
+#ifdef HAVE_INOTIFY
+ /* During an invalidation we try to reload the traced
+ file watches. This allows the user to re-sync if
+ inotify events were lost. Similar to what we do during
+ pruning. */
+ install_watches (runp);
+#endif
+ if (runp->call_res_init)
+ {
+ res_init ();
+ break;
+ }
+ runp = runp->next;
+ }
break;
}
if (dbs[number].enabled)
{
- pthread_mutex_lock (&dbs[number].prune_lock);
+ pthread_mutex_lock (&dbs[number].prune_run_lock);
prune_cache (&dbs[number], LONG_MAX, fd);
- pthread_mutex_unlock (&dbs[number].prune_lock);
+ pthread_mutex_unlock (&dbs[number].prune_run_lock);
}
else
{
cmsg->cmsg_type = SCM_RIGHTS;
cmsg->cmsg_len = CMSG_LEN (sizeof (int));
- *(int *) CMSG_DATA (cmsg) = db->ro_fd;
+ int *ip = (int *) CMSG_DATA (cmsg);
+ *ip = db->ro_fd;
msg.msg_controllen = cmsg->cmsg_len;
#endif
(void) TEMP_FAILURE_RETRY (sendmsg (fd, &msg, MSG_NOSIGNAL));
- if (__builtin_expect (debug_level > 0, 0))
+ if (__glibc_unlikely (debug_level > 0))
dbg_log (_("provide access to FD %d, for %s"), db->ro_fd, key);
}
#endif /* SCM_RIGHTS */
}
/* Is this service enabled? */
- if (__builtin_expect (!db->enabled, 0))
+ if (__glibc_unlikely (!db->enabled))
{
/* No, sent the prepared record. */
if (TEMP_FAILURE_RETRY (send (fd, db->disabled_iov->iov_base,
}
/* Be sure we can read the data. */
- if (__builtin_expect (pthread_rwlock_tryrdlock (&db->lock) != 0, 0))
+ if (__glibc_unlikely (pthread_rwlock_tryrdlock (&db->lock) != 0))
{
++db->head->rdlockdelayed;
pthread_rwlock_rdlock (&db->lock);
ssize_t nwritten;
#ifdef HAVE_SENDFILE
- if (__builtin_expect (db->mmap_used, 1))
+ if (__glibc_likely (db->mmap_used))
{
assert (db->wr_fd != -1);
assert ((char *) cached->data > (char *) db->data);
addservbyport (db, fd, req, key, uid);
break;
+ case GETNETGRENT:
+ addgetnetgrent (db, fd, req, key, uid);
+ break;
+
+ case INNETGR:
+ addinnetgr (db, fd, req, key, uid);
+ break;
+
case GETSTAT:
case SHUTDOWN:
case INVALIDATE:
case GETFDGR:
case GETFDHST:
case GETFDSERV:
+ case GETFDNETGR:
#ifdef SCM_RIGHTS
send_ro_fd (reqinfo[req->type].db, key, fd);
#endif
cannot change to old GID: %s; disabling paranoia mode"),
strerror (errno));
- setuid (server_uid);
+ ignore_value (setuid (server_uid));
paranoia = 0;
return;
}
if (server_user != NULL)
{
- setuid (server_uid);
- setgid (server_gid);
+ ignore_value (setuid (server_uid));
+ ignore_value (setgid (server_gid));
}
paranoia = 0;
return;
}
/* The preparations are done. */
+#ifdef PATH_MAX
+ char pathbuf[PATH_MAX];
+#else
+ char pathbuf[256];
+#endif
+ /* Try to exec the real nscd program so the process name (as reported
+ in /proc/PID/status) will be 'nscd', but fall back to /proc/self/exe
+ if readlink or the exec with the result of the readlink call fails. */
+ ssize_t n = readlink ("/proc/self/exe", pathbuf, sizeof (pathbuf) - 1);
+ if (n != -1)
+ {
+ pathbuf[n] = '\0';
+ execv (pathbuf, argv);
+ }
execv ("/proc/self/exe", argv);
/* If we come here, we will never be able to re-exec. */
if (server_user != NULL)
{
- setuid (server_uid);
- setgid (server_gid);
+ ignore_value (setuid (server_uid));
+ ignore_value (setgid (server_gid));
}
if (chdir ("/") != 0)
dbg_log (_("cannot change current working directory to \"/\": %s"),
dbs[my_number].head->timestamp = now;
struct timespec prune_ts;
- if (__builtin_expect (clock_gettime (timeout_clock, &prune_ts) == -1, 0))
+ if (__glibc_unlikely (clock_gettime (timeout_clock, &prune_ts) == -1))
/* Should never happen. */
abort ();
dbs[my_number].wakeup_time = now + CACHE_PRUNE_INTERVAL + my_number;
pthread_mutex_t *prune_lock = &dbs[my_number].prune_lock;
+ pthread_mutex_t *prune_run_lock = &dbs[my_number].prune_run_lock;
pthread_cond_t *prune_cond = &dbs[my_number].prune_cond;
pthread_mutex_lock (prune_lock);
pruning we want to know about it. Therefore set the
timeout to the maximum. It will be descreased when adding
new entries to the cache, if necessary. */
- if (sizeof (time_t) == sizeof (long int))
- dbs[my_number].wakeup_time = LONG_MAX;
- else
- dbs[my_number].wakeup_time = INT_MAX;
+ dbs[my_number].wakeup_time = MAX_TIMEOUT_VALUE;
/* Unconditionally reset the flag. */
time_t prune_now = dbs[my_number].clear_cache ? LONG_MAX : now;
pthread_mutex_unlock (prune_lock);
+ /* We use a separate lock for running the prune function (instead
+ of keeping prune_lock locked) because this enables concurrent
+ invocations of cache_add which might modify the timeout value. */
+ pthread_mutex_lock (prune_run_lock);
next_wait = prune_cache (&dbs[my_number], prune_now, -1);
+ pthread_mutex_unlock (prune_run_lock);
next_wait = MAX (next_wait, CACHE_PRUNE_INTERVAL);
/* If clients cannot determine for sure whether nscd is running
we need to wake up occasionally to update the timestamp.
Wait 90% of the update period. */
#define UPDATE_MAPPING_TIMEOUT (MAPPING_TIMEOUT * 9 / 10)
- if (__builtin_expect (! dont_need_update, 0))
+ if (__glibc_unlikely (! dont_need_update))
{
next_wait = MIN (UPDATE_MAPPING_TIMEOUT, next_wait);
dbs[my_number].head->timestamp = now;
/* This is the main loop. It is replicated in different threads but
- the the use of the ready list makes sure only one thread handles an
+ the use of the ready list makes sure only one thread handles an
incoming connection. */
static void *
__attribute__ ((__noreturn__))
{
char buf[256];
- /* Initialize the memory-in-flight list. */
- for (enum in_flight idx = 0; idx < IDX_last; ++idx)
- mem_in_flight.block[idx].dbidx = -1;
- /* And queue this threads structure. */
- do
- mem_in_flight.next = mem_in_flight_list;
- while (atomic_compare_and_exchange_bool_acq (&mem_in_flight_list,
- &mem_in_flight,
- mem_in_flight.next) != 0);
-
/* Initial locking. */
pthread_mutex_lock (&readylist_lock);
/* We are done with the list. */
pthread_mutex_unlock (&readylist_lock);
-#ifndef __ASSUME_SOCK_CLOEXEC
- if (have_sock_cloexec < 0)
+#ifndef __ASSUME_ACCEPT4
+ if (have_accept4 < 0)
{
/* We do not want to block on a short read or so. */
int fl = fcntl (fd, F_GETFL);
#ifdef SO_PEERCRED
pid_t pid = 0;
- if (__builtin_expect (debug_level > 0, 0))
+ if (__glibc_unlikely (debug_level > 0))
{
struct ucred caller;
socklen_t optlen = sizeof (caller);
else
{
/* Get the key. */
- char keybuf[MAXKEYLEN];
+ char keybuf[MAXKEYLEN + 1];
if (__builtin_expect (TEMP_FAILURE_RETRY (read (fd, keybuf,
req.key_len))
strerror_r (errno, buf, sizeof (buf)));
goto close_and_out;
}
+ keybuf[req.key_len] = '\0';
if (__builtin_expect (debug_level, 0) > 0)
{
/* One more thread available. */
++nready;
}
+ /* NOTREACHED */
}
}
bool do_signal = true;
- if (__builtin_expect (nready == 0, 0))
+ if (__glibc_unlikely (nready == 0))
{
++client_queued;
do_signal = false;
/* Check whether restarting should happen. */
-static inline int
+static bool
restart_p (time_t now)
{
return (paranoia && readylist == NULL && nready == nthreads
/* Array for times a connection was accepted. */
static time_t *starttime;
+#ifdef HAVE_INOTIFY
+/* Inotify event for changed file. */
+union __inev
+{
+ struct inotify_event i;
+# ifndef PATH_MAX
+# define PATH_MAX 1024
+# endif
+ char buf[sizeof (struct inotify_event) + PATH_MAX];
+};
+
+/* Returns 0 if the file is there otherwise -1. */
+int
+check_file (struct traced_file *finfo)
+{
+ struct stat64 st;
+ /* We could check mtime and if different re-add
+ the watches, and invalidate the database, but we
+ don't because we are called from inotify_check_files
+ which should be doing that work. If sufficient inotify
+ events were lost then the next pruning or invalidation
+ will do the stat and mtime check. We don't do it here to
+ keep the logic simple. */
+ if (stat64 (finfo->fname, &st) < 0)
+ return -1;
+ return 0;
+}
+
+/* Process the inotify event in INEV. If the event matches any of the files
+ registered with a database then mark that database as requiring its cache
+ to be cleared. We indicate the cache needs clearing by setting
+ TO_CLEAR[DBCNT] to true for the matching database. */
+static void
+inotify_check_files (bool *to_clear, union __inev *inev)
+{
+ /* Check which of the files changed. */
+ for (size_t dbcnt = 0; dbcnt < lastdb; ++dbcnt)
+ {
+ struct traced_file *finfo = dbs[dbcnt].traced_files;
+
+ while (finfo != NULL)
+ {
+ /* The configuration file was moved or deleted.
+ We stop watching it at that point, and reinitialize. */
+ if (finfo->inotify_descr[TRACED_FILE] == inev->i.wd
+ && ((inev->i.mask & IN_MOVE_SELF)
+ || (inev->i.mask & IN_DELETE_SELF)
+ || (inev->i.mask & IN_IGNORED)))
+ {
+ int ret;
+ bool moved = (inev->i.mask & IN_MOVE_SELF) != 0;
+
+ if (check_file (finfo) == 0)
+ {
+ dbg_log (_("ignored inotify event for `%s` (file exists)"),
+ finfo->fname);
+ return;
+ }
+
+ dbg_log (_("monitored file `%s` was %s, removing watch"),
+ finfo->fname, moved ? "moved" : "deleted");
+ /* File was moved out, remove the watch. Watches are
+ automatically removed when the file is deleted. */
+ if (moved)
+ {
+ ret = inotify_rm_watch (inotify_fd, inev->i.wd);
+ if (ret < 0)
+ dbg_log (_("failed to remove file watch `%s`: %s"),
+ finfo->fname, strerror (errno));
+ }
+ finfo->inotify_descr[TRACED_FILE] = -1;
+ to_clear[dbcnt] = true;
+ if (finfo->call_res_init)
+ res_init ();
+ return;
+ }
+ /* The configuration file was open for writing and has just closed.
+ We reset the cache and reinitialize. */
+ if (finfo->inotify_descr[TRACED_FILE] == inev->i.wd
+ && inev->i.mask & IN_CLOSE_WRITE)
+ {
+ /* Mark cache as needing to be cleared and reinitialize. */
+ dbg_log (_("monitored file `%s` was written to"), finfo->fname);
+ to_clear[dbcnt] = true;
+ if (finfo->call_res_init)
+ res_init ();
+ return;
+ }
+ /* The parent directory was moved or deleted. We trigger one last
+ invalidation. At the next pruning or invalidation we may add
+ this watch back if the file is present again. */
+ if (finfo->inotify_descr[TRACED_DIR] == inev->i.wd
+ && ((inev->i.mask & IN_DELETE_SELF)
+ || (inev->i.mask & IN_MOVE_SELF)
+ || (inev->i.mask & IN_IGNORED)))
+ {
+ bool moved = (inev->i.mask & IN_MOVE_SELF) != 0;
+ /* The directory watch may have already been removed
+ but we don't know so we just remove it again and
+ ignore the error. Then we remove the file watch.
+ Note: watches are automatically removed for deleted
+ files. */
+ if (moved)
+ inotify_rm_watch (inotify_fd, inev->i.wd);
+ if (finfo->inotify_descr[TRACED_FILE] != -1)
+ {
+ dbg_log (_("monitored parent directory `%s` was %s, removing watch on `%s`"),
+ finfo->dname, moved ? "moved" : "deleted", finfo->fname);
+ if (inotify_rm_watch (inotify_fd, finfo->inotify_descr[TRACED_FILE]) < 0)
+ dbg_log (_("failed to remove file watch `%s`: %s"),
+ finfo->dname, strerror (errno));
+ }
+ finfo->inotify_descr[TRACED_FILE] = -1;
+ finfo->inotify_descr[TRACED_DIR] = -1;
+ to_clear[dbcnt] = true;
+ if (finfo->call_res_init)
+ res_init ();
+ /* Continue to the next entry since this might be the
+ parent directory for multiple registered files and
+ we want to remove watches for all registered files. */
+ continue;
+ }
+ /* The parent directory had a create or moved to event. */
+ if (finfo->inotify_descr[TRACED_DIR] == inev->i.wd
+ && ((inev->i.mask & IN_MOVED_TO)
+ || (inev->i.mask & IN_CREATE))
+ && strcmp (inev->i.name, finfo->sfname) == 0)
+ {
+ /* We detected a directory change. We look for the creation
+ of the file we are tracking or the move of the same file
+ into the directory. */
+ int ret;
+ dbg_log (_("monitored file `%s` was %s, adding watch"),
+ finfo->fname,
+ inev->i.mask & IN_CREATE ? "created" : "moved into place");
+ /* File was moved in or created. Regenerate the watch. */
+ if (finfo->inotify_descr[TRACED_FILE] != -1)
+ inotify_rm_watch (inotify_fd,
+ finfo->inotify_descr[TRACED_FILE]);
+
+ ret = inotify_add_watch (inotify_fd,
+ finfo->fname,
+ TRACED_FILE_MASK);
+ if (ret < 0)
+ dbg_log (_("failed to add file watch `%s`: %s"),
+ finfo->fname, strerror (errno));
+
+ finfo->inotify_descr[TRACED_FILE] = ret;
+
+ /* The file is new or moved so mark cache as needing to
+ be cleared and reinitialize. */
+ to_clear[dbcnt] = true;
+ if (finfo->call_res_init)
+ res_init ();
+
+ /* Done re-adding the watch. Don't return, we may still
+ have other files in this same directory, same watch
+ descriptor, and need to process them. */
+ }
+ /* Other events are ignored, and we move on to the next file. */
+ finfo = finfo->next;
+ }
+ }
+}
+
+/* If an entry in the array of booleans TO_CLEAR is TRUE then clear the cache
+ for the associated database, otherwise do nothing. The TO_CLEAR array must
+ have LASTDB entries. */
+static inline void
+clear_db_cache (bool *to_clear)
+{
+ for (size_t dbcnt = 0; dbcnt < lastdb; ++dbcnt)
+ if (to_clear[dbcnt])
+ {
+ pthread_mutex_lock (&dbs[dbcnt].prune_lock);
+ dbs[dbcnt].clear_cache = 1;
+ pthread_mutex_unlock (&dbs[dbcnt].prune_lock);
+ pthread_cond_signal (&dbs[dbcnt].prune_cond);
+ }
+}
+
+int
+handle_inotify_events (void)
+{
+ bool to_clear[lastdb] = { false, };
+ union __inev inev;
+
+ /* Read all inotify events for files registered via
+ register_traced_file(). */
+ while (1)
+ {
+ /* Potentially read multiple events into buf. */
+ ssize_t nb = TEMP_FAILURE_RETRY (read (inotify_fd,
+ &inev.buf,
+ sizeof (inev)));
+ if (nb < (ssize_t) sizeof (struct inotify_event))
+ {
+ /* Not even 1 event. */
+ if (__glibc_unlikely (nb == -1 && errno != EAGAIN))
+ return -1;
+ /* Done reading events that are ready. */
+ break;
+ }
+ /* Process all events. The normal inotify interface delivers
+ complete events on a read and never a partial event. */
+ char *eptr = &inev.buf[0];
+ ssize_t count;
+ while (1)
+ {
+ /* Check which of the files changed. */
+ inotify_check_files (to_clear, &inev);
+ count = sizeof (struct inotify_event) + inev.i.len;
+ eptr += count;
+ nb -= count;
+ if (nb >= (ssize_t) sizeof (struct inotify_event))
+ memcpy (&inev, eptr, nb);
+ else
+ break;
+ }
+ continue;
+ }
+ /* Actually perform the cache clearing. */
+ clear_db_cache (to_clear);
+ return 0;
+}
+
+#endif
static void
__attribute__ ((__noreturn__))
}
#endif
+#ifdef HAVE_NETLINK
+ size_t idx_nl_status_fd = 0;
+ if (nl_status_fd != -1)
+ {
+ idx_nl_status_fd = nused;
+ conns[nused].fd = nl_status_fd;
+ conns[nused].events = POLLRDNORM;
+ ++nused;
+ firstfree = nused;
+ }
+#endif
+
while (1)
{
/* Wait for any event. We wait at most a couple of seconds so
/* We have a new incoming connection. Accept the connection. */
int fd;
-#ifndef __ASSUME_PACCEPT
+#ifndef __ASSUME_ACCEPT4
fd = -1;
- if (have_paccept >= 0)
+ if (have_accept4 >= 0)
#endif
{
- fd = TEMP_FAILURE_RETRY (paccept (sock, NULL, NULL, NULL,
+ fd = TEMP_FAILURE_RETRY (accept4 (sock, NULL, NULL,
SOCK_NONBLOCK));
-#ifndef __ASSUME_PACCEPT
- if (have_paccept == 0)
- have_paccept = fd != -1 || errno != ENOSYS ? 1 : -1;
+#ifndef __ASSUME_ACCEPT4
+ if (have_accept4 == 0)
+ have_accept4 = fd != -1 || errno != ENOSYS ? 1 : -1;
#endif
}
-#ifndef __ASSUME_PACCEPT
- if (have_paccept < 0)
+#ifndef __ASSUME_ACCEPT4
+ if (have_accept4 < 0)
fd = TEMP_FAILURE_RETRY (accept (sock, NULL, NULL));
#endif
{
if (conns[1].revents != 0)
{
- bool to_clear[lastdb] = { false, };
- union
- {
- struct inotify_event i;
- char buf[100];
- } inev;
-
- while (1)
+ int ret;
+ ret = handle_inotify_events ();
+ if (ret == -1)
{
- ssize_t nb = TEMP_FAILURE_RETRY (read (inotify_fd, &inev,
- sizeof (inev)));
- if (nb < (ssize_t) sizeof (struct inotify_event))
- {
- if (nb == -1)
- {
- /* Something went wrong when reading the inotify
- data. Better disable inotify. */
- conns[1].fd = -1;
- firstfree = 1;
- if (nused == 2)
- nused = 1;
- close (inotify_fd);
- inotify_fd = -1;
- dbg_log (_("disabled inotify after read error"));
- }
- break;
- }
-
- /* Check which of the files changed. */
- for (size_t dbcnt = 0; dbcnt < lastdb; ++dbcnt)
- if (inev.i.wd == dbs[dbcnt].inotify_descr)
- {
- to_clear[dbcnt] = true;
- goto next;
- }
-
- if (inev.i.wd == resolv_conf_descr)
- {
- res_init ();
- to_clear[hstdb] = true;
- }
- next:;
+ /* Something went wrong when reading the inotify
+ data. Better disable inotify. */
+ dbg_log (_("disabled inotify-based monitoring after read error %d"), errno);
+ conns[1].fd = -1;
+ firstfree = 1;
+ if (nused == 2)
+ nused = 1;
+ close (inotify_fd);
+ inotify_fd = -1;
}
-
- /* Actually perform the cache clearing. */
- for (size_t dbcnt = 0; dbcnt < lastdb; ++dbcnt)
- if (to_clear[dbcnt])
- {
- pthread_mutex_lock (&dbs[dbcnt].prune_lock);
- dbs[dbcnt].clear_cache = 1;
- pthread_mutex_unlock (&dbs[dbcnt].prune_lock);
- pthread_cond_signal (&dbs[dbcnt].prune_cond);
- }
-
--n;
}
}
#endif
+#ifdef HAVE_NETLINK
+ if (idx_nl_status_fd != 0 && conns[idx_nl_status_fd].revents != 0)
+ {
+ char buf[4096];
+ /* Read all the data. We do not interpret it here. */
+ while (TEMP_FAILURE_RETRY (read (nl_status_fd, buf,
+ sizeof (buf))) != -1)
+ ;
+
+ dbs[hstdb].head->extra_data[NSCD_HST_IDX_CONF_TIMESTAMP]
+ = __bump_nl_timestamp ();
+ }
+#endif
+
for (size_t cnt = first; cnt < nused && n > 0; ++cnt)
if (conns[cnt].revents != 0)
{
/* We cannot use epoll. */
return;
-#ifdef HAVE_INOTIFY
+# ifdef HAVE_INOTIFY
if (inotify_fd != -1)
{
ev.events = EPOLLRDNORM;
return;
nused = 2;
}
-#endif
+# endif
+
+# ifdef HAVE_NETLINK
+ if (nl_status_fd != -1)
+ {
+ ev.events = EPOLLRDNORM;
+ ev.data.fd = nl_status_fd;
+ if (epoll_ctl (efd, EPOLL_CTL_ADD, nl_status_fd, &ev) == -1)
+ /* We cannot use epoll. */
+ return;
+ }
+# endif
while (1)
{
if (revs[cnt].data.fd == sock)
{
/* A new connection. */
- int fd = TEMP_FAILURE_RETRY (accept (sock, NULL, NULL));
+ int fd;
+# ifndef __ASSUME_ACCEPT4
+ fd = -1;
+ if (have_accept4 >= 0)
+# endif
+ {
+ fd = TEMP_FAILURE_RETRY (accept4 (sock, NULL, NULL,
+ SOCK_NONBLOCK));
+# ifndef __ASSUME_ACCEPT4
+ if (have_accept4 == 0)
+ have_accept4 = fd != -1 || errno != ENOSYS ? 1 : -1;
+# endif
+ }
+# ifndef __ASSUME_ACCEPT4
+ if (have_accept4 < 0)
+ fd = TEMP_FAILURE_RETRY (accept (sock, NULL, NULL));
+# endif
+
+ /* Use the descriptor if we have not reached the limit. */
if (fd >= 0)
{
/* Try to add the new descriptor. */
}
}
}
-#ifdef HAVE_INOTIFY
+# ifdef HAVE_INOTIFY
else if (revs[cnt].data.fd == inotify_fd)
{
- bool to_clear[lastdb] = { false, };
- union
- {
- struct inotify_event i;
- char buf[100];
- } inev;
-
- while (1)
+ int ret;
+ ret = handle_inotify_events ();
+ if (ret == -1)
{
- ssize_t nb = TEMP_FAILURE_RETRY (read (inotify_fd, &inev,
- sizeof (inev)));
- if (nb < (ssize_t) sizeof (struct inotify_event))
- {
- if (nb == -1)
- {
- /* Something went wrong when reading the inotify
- data. Better disable inotify. */
- (void) epoll_ctl (efd, EPOLL_CTL_DEL, inotify_fd,
- NULL);
- close (inotify_fd);
- inotify_fd = -1;
- dbg_log (_("disabled inotify after read error"));
- }
- break;
- }
-
- /* Check which of the files changed. */
- for (size_t dbcnt = 0; dbcnt < lastdb; ++dbcnt)
- if (inev.i.wd == dbs[dbcnt].inotify_descr)
- {
- to_clear[dbcnt] = true;
- goto next;
- }
-
- if (inev.i.wd == resolv_conf_descr)
- {
- res_init ();
- to_clear[hstdb] = true;
- }
- next:;
+ /* Something went wrong when reading the inotify
+ data. Better disable inotify. */
+ dbg_log (_("disabled inotify-based monitoring after read error %d"), errno);
+ (void) epoll_ctl (efd, EPOLL_CTL_DEL, inotify_fd, NULL);
+ close (inotify_fd);
+ inotify_fd = -1;
+ break;
}
+ }
+# endif
+# ifdef HAVE_NETLINK
+ else if (revs[cnt].data.fd == nl_status_fd)
+ {
+ char buf[4096];
+ /* Read all the data. We do not interpret it here. */
+ while (TEMP_FAILURE_RETRY (read (nl_status_fd, buf,
+ sizeof (buf))) != -1)
+ ;
- /* Actually perform the cache clearing. */
- for (size_t dbcnt = 0; dbcnt < lastdb; ++dbcnt)
- if (to_clear[dbcnt])
- {
- pthread_mutex_lock (&dbs[dbcnt].prune_lock);
- dbs[dbcnt].clear_cache = 1;
- pthread_mutex_unlock (&dbs[dbcnt].prune_lock);
- pthread_cond_signal (&dbs[dbcnt].prune_cond);
- }
+ __bump_nl_timestamp ();
}
-#endif
+# endif
else
{
/* Remove the descriptor from the epoll descriptor. */
no reply in too long of a time. */
time_t laststart = now - ACCEPT_TIMEOUT;
assert (starttime[sock] == 0);
+# ifdef HAVE_INOTIFY
assert (inotify_fd == -1 || starttime[inotify_fd] == 0);
+# endif
+ assert (nl_status_fd == -1 || starttime[nl_status_fd] == 0);
for (int cnt = highest; cnt > STDERR_FILENO; --cnt)
if (starttime[cnt] != 0 && starttime[cnt] < laststart)
{
if (pthread_cond_init (&dbs[i].prune_cond, &condattr) != 0)
{
dbg_log (_("could not initialize conditional variable"));
- exit (1);
+ do_exit (1, 0, NULL);
}
pthread_t th;
&& pthread_create (&th, &attr, nscd_run_prune, (void *) i) != 0)
{
dbg_log (_("could not start clean-up thread; terminating"));
- exit (1);
+ do_exit (1, 0, NULL);
}
}
if (i == 0)
{
dbg_log (_("could not start any worker thread; terminating"));
- exit (1);
+ do_exit (1, 0, NULL);
}
break;
}
}
+ /* Now it is safe to let the parent know that we're doing fine and it can
+ exit. */
+ notify_parent (0);
+
/* Determine how much room for descriptors we should initially
allocate. This might need to change later if we cap the number
with MAXCONN. */
if (pwd == NULL)
{
dbg_log (_("Failed to run nscd as user '%s'"), server_user);
- error (EXIT_FAILURE, 0, _("Failed to run nscd as user '%s'"),
- server_user);
+ do_exit (EXIT_FAILURE, 0,
+ _("Failed to run nscd as user '%s'"), server_user);
}
server_uid = pwd->pw_uid;
{
/* This really must never happen. */
dbg_log (_("Failed to run nscd as user '%s'"), server_user);
- error (EXIT_FAILURE, errno, _("initial getgrouplist failed"));
+ do_exit (EXIT_FAILURE, errno,
+ _("initial getgrouplist failed"));
}
server_groups = (gid_t *) xmalloc (server_ngroups * sizeof (gid_t));
== -1)
{
dbg_log (_("Failed to run nscd as user '%s'"), server_user);
- error (EXIT_FAILURE, errno, _("getgrouplist failed"));
+ do_exit (EXIT_FAILURE, errno, _("getgrouplist failed"));
}
}
if (setgroups (server_ngroups, server_groups) == -1)
{
dbg_log (_("Failed to run nscd as user '%s'"), server_user);
- error (EXIT_FAILURE, errno, _("setgroups failed"));
+ do_exit (EXIT_FAILURE, errno, _("setgroups failed"));
}
int res;
if (res == -1)
{
dbg_log (_("Failed to run nscd as user '%s'"), server_user);
- perror ("setgid");
- exit (4);
+ do_exit (4, errno, "setgid");
}
if (paranoia)
if (res == -1)
{
dbg_log (_("Failed to run nscd as user '%s'"), server_user);
- perror ("setuid");
- exit (4);
+ do_exit (4, errno, "setuid");
}
#if defined HAVE_LIBAUDIT && defined HAVE_LIBCAP