1 /* Inner loops of cache daemon.
2 Copyright (C) 1998-2003, 2004, 2005 Free Software Foundation, Inc.
3 This file is part of the GNU C Library.
4 Contributed by Ulrich Drepper <drepper@cygnus.com>, 1998.
6 The GNU C Library is free software; you can redistribute it and/or
7 modify it under the terms of the GNU Lesser General Public
8 License as published by the Free Software Foundation; either
9 version 2.1 of the License, or (at your option) any later version.
11 The GNU C Library is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
16 You should have received a copy of the GNU Lesser General Public
17 License along with the GNU C Library; if not, write to the Free
18 Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
36 #include <arpa/inet.h>
38 # include <sys/epoll.h>
41 #include <sys/param.h>
43 #include <sys/socket.h>
46 #include <gnu/lib-names.h>
53 /* Number of bytes of data we initially reserve for each hash table bucket. */
54 #define DEFAULT_DATASIZE_PER_BUCKET 1024
57 /* Wrapper functions with error checking for standard functions. */
58 extern void *xmalloc (size_t n
);
59 extern void *xcalloc (size_t n
, size_t s
);
60 extern void *xrealloc (void *o
, size_t n
);
62 /* Support to run nscd as an unprivileged user */
63 const char *server_user
;
64 static uid_t server_uid
;
65 static gid_t server_gid
;
66 const char *stat_user
;
68 static gid_t
*server_groups
;
72 static int server_ngroups
;
74 static pthread_attr_t attr
;
76 static void begin_drop_privileges (void);
77 static void finish_drop_privileges (void);
79 /* Map request type to a string. */
80 const char *serv2str
[LASTREQ
] =
82 [GETPWBYNAME
] = "GETPWBYNAME",
83 [GETPWBYUID
] = "GETPWBYUID",
84 [GETGRBYNAME
] = "GETGRBYNAME",
85 [GETGRBYGID
] = "GETGRBYGID",
86 [GETHOSTBYNAME
] = "GETHOSTBYNAME",
87 [GETHOSTBYNAMEv6
] = "GETHOSTBYNAMEv6",
88 [GETHOSTBYADDR
] = "GETHOSTBYADDR",
89 [GETHOSTBYADDRv6
] = "GETHOSTBYADDRv6",
90 [SHUTDOWN
] = "SHUTDOWN",
91 [GETSTAT
] = "GETSTAT",
92 [INVALIDATE
] = "INVALIDATE",
93 [GETFDPW
] = "GETFDPW",
94 [GETFDGR
] = "GETFDGR",
95 [GETFDHST
] = "GETFDHST",
97 [INITGROUPS
] = "INITGROUPS"
100 /* The control data structures for the services. */
101 struct database_dyn dbs
[lastdb
] =
104 .lock
= PTHREAD_RWLOCK_WRITER_NONRECURSIVE_INITIALIZER_NP
,
109 .filename
= "/etc/passwd",
110 .db_filename
= _PATH_NSCD_PASSWD_DB
,
111 .disabled_iov
= &pwd_iov_disabled
,
119 .lock
= PTHREAD_RWLOCK_WRITER_NONRECURSIVE_INITIALIZER_NP
,
124 .filename
= "/etc/group",
125 .db_filename
= _PATH_NSCD_GROUP_DB
,
126 .disabled_iov
= &grp_iov_disabled
,
134 .lock
= PTHREAD_RWLOCK_WRITER_NONRECURSIVE_INITIALIZER_NP
,
139 .filename
= "/etc/hosts",
140 .db_filename
= _PATH_NSCD_HOSTS_DB
,
141 .disabled_iov
= &hst_iov_disabled
,
151 /* Mapping of request type to database. */
152 static struct database_dyn
*const serv2db
[LASTREQ
] =
154 [GETPWBYNAME
] = &dbs
[pwddb
],
155 [GETPWBYUID
] = &dbs
[pwddb
],
156 [GETGRBYNAME
] = &dbs
[grpdb
],
157 [GETGRBYGID
] = &dbs
[grpdb
],
158 [GETHOSTBYNAME
] = &dbs
[hstdb
],
159 [GETHOSTBYNAMEv6
] = &dbs
[hstdb
],
160 [GETHOSTBYADDR
] = &dbs
[hstdb
],
161 [GETHOSTBYADDRv6
] = &dbs
[hstdb
],
162 [GETFDPW
] = &dbs
[pwddb
],
163 [GETFDGR
] = &dbs
[grpdb
],
164 [GETFDHST
] = &dbs
[hstdb
],
165 [GETAI
] = &dbs
[hstdb
],
166 [INITGROUPS
] = &dbs
[grpdb
]
170 /* Number of seconds between two cache pruning runs. */
171 #define CACHE_PRUNE_INTERVAL 15
174 /* Initial number of threads to use. */
176 /* Maximum number of threads to use. */
177 int max_nthreads
= 32;
179 /* Socket for incoming connections. */
182 /* Number of times clients had to wait. */
183 unsigned long int client_queued
;
187 writeall (int fd
, const void *buf
, size_t len
)
193 ret
= TEMP_FAILURE_RETRY (write (fd
, buf
, n
));
196 buf
= (const char *) buf
+ ret
;
200 return ret
< 0 ? ret
: len
- n
;
204 /* Initialize database information structures. */
208 /* Secure mode and unprivileged mode are incompatible */
209 if (server_user
!= NULL
&& secure_in_use
)
211 dbg_log (_("Cannot run nscd in secure mode as unprivileged user"));
215 /* Look up unprivileged uid/gid/groups before we start listening on the
217 if (server_user
!= NULL
)
218 begin_drop_privileges ();
221 /* No configuration for this value, assume a default. */
222 nthreads
= 2 * lastdb
;
224 for (size_t cnt
= 0; cnt
< lastdb
; ++cnt
)
225 if (dbs
[cnt
].enabled
)
227 pthread_rwlock_init (&dbs
[cnt
].lock
, NULL
);
228 pthread_mutex_init (&dbs
[cnt
].memlock
, NULL
);
230 if (dbs
[cnt
].persistent
)
232 /* Try to open the appropriate file on disk. */
233 int fd
= open (dbs
[cnt
].db_filename
, O_RDWR
);
239 struct database_pers_head head
;
240 ssize_t n
= TEMP_FAILURE_RETRY (read (fd
, &head
,
242 if (n
!= sizeof (head
) || fstat64 (fd
, &st
) != 0)
245 dbg_log (_("invalid persistent database file \"%s\": %s"),
246 dbs
[cnt
].db_filename
, strerror (errno
));
247 dbs
[cnt
].persistent
= 0;
249 else if (head
.module
== 0 && head
.data_size
== 0)
251 /* The file has been created, but the head has not been
252 initialized yet. Remove the old file. */
253 unlink (dbs
[cnt
].db_filename
);
255 else if (head
.header_size
!= (int) sizeof (head
))
257 dbg_log (_("invalid persistent database file \"%s\": %s"),
258 dbs
[cnt
].db_filename
,
259 _("header size does not match"));
260 dbs
[cnt
].persistent
= 0;
262 else if ((total
= (sizeof (head
)
263 + roundup (head
.module
* sizeof (ref_t
),
268 dbg_log (_("invalid persistent database file \"%s\": %s"),
269 dbs
[cnt
].db_filename
,
270 _("file size does not match"));
271 dbs
[cnt
].persistent
= 0;
273 else if ((mem
= mmap (NULL
, total
, PROT_READ
| PROT_WRITE
,
274 MAP_SHARED
, fd
, 0)) == MAP_FAILED
)
278 /* Success. We have the database. */
280 dbs
[cnt
].memsize
= total
;
281 dbs
[cnt
].data
= (char *)
282 &dbs
[cnt
].head
->array
[roundup (dbs
[cnt
].head
->module
,
283 ALIGN
/ sizeof (ref_t
))];
284 dbs
[cnt
].mmap_used
= true;
286 if (dbs
[cnt
].suggested_module
> head
.module
)
287 dbg_log (_("suggested size of table for database %s larger than the persistent database's table"),
292 /* We also need a read-only descriptor. */
295 dbs
[cnt
].ro_fd
= open (dbs
[cnt
].db_filename
, O_RDONLY
);
296 if (dbs
[cnt
].ro_fd
== -1)
298 cannot create read-only descriptor for \"%s\"; no mmap"),
299 dbs
[cnt
].db_filename
);
302 // XXX Shall we test whether the descriptors actually
303 // XXX point to the same file?
306 /* Close the file descriptors in case something went
307 wrong in which case the variable have not been
314 if (dbs
[cnt
].head
== NULL
)
316 /* No database loaded. Allocate the data structure,
318 struct database_pers_head head
;
319 size_t total
= (sizeof (head
)
320 + roundup (dbs
[cnt
].suggested_module
321 * sizeof (ref_t
), ALIGN
)
322 + (dbs
[cnt
].suggested_module
323 * DEFAULT_DATASIZE_PER_BUCKET
));
325 /* Try to create the database. If we do not need a
326 persistent database create a temporary file. */
329 if (dbs
[cnt
].persistent
)
331 fd
= open (dbs
[cnt
].db_filename
,
332 O_RDWR
| O_CREAT
| O_EXCL
| O_TRUNC
,
334 if (fd
!= -1 && dbs
[cnt
].shared
)
335 ro_fd
= open (dbs
[cnt
].db_filename
, O_RDONLY
);
339 char fname
[] = _PATH_NSCD_XYZ_DB_TMP
;
340 fd
= mkstemp (fname
);
342 /* We do not need the file name anymore after we
343 opened another file descriptor in read-only mode. */
347 ro_fd
= open (fname
, O_RDONLY
);
357 dbg_log (_("database for %s corrupted or simultaneously used; remove %s manually if necessary and restart"),
358 dbnames
[cnt
], dbs
[cnt
].db_filename
);
359 // XXX Correct way to terminate?
363 if (dbs
[cnt
].persistent
)
364 dbg_log (_("cannot create %s; no persistent database used"),
365 dbs
[cnt
].db_filename
);
367 dbg_log (_("cannot create %s; no sharing possible"),
368 dbs
[cnt
].db_filename
);
370 dbs
[cnt
].persistent
= 0;
371 // XXX remember: no mmap
375 /* Tell the user if we could not create the read-only
377 if (ro_fd
== -1 && dbs
[cnt
].shared
)
379 cannot create read-only descriptor for \"%s\"; no mmap"),
380 dbs
[cnt
].db_filename
);
382 /* Before we create the header, initialiye the hash
383 table. So that if we get interrupted if writing
384 the header we can recognize a partially initialized
386 size_t ps
= sysconf (_SC_PAGESIZE
);
388 assert (~ENDREF
== 0);
389 memset (tmpbuf
, '\xff', ps
);
391 size_t remaining
= dbs
[cnt
].suggested_module
* sizeof (ref_t
);
392 off_t offset
= sizeof (head
);
395 if (offset
% ps
!= 0)
397 towrite
= MIN (remaining
, ps
- (offset
% ps
));
398 pwrite (fd
, tmpbuf
, towrite
, offset
);
400 remaining
-= towrite
;
403 while (remaining
> ps
)
405 pwrite (fd
, tmpbuf
, ps
, offset
);
411 pwrite (fd
, tmpbuf
, remaining
, offset
);
413 /* Create the header of the file. */
414 struct database_pers_head head
=
416 .version
= DB_VERSION
,
417 .header_size
= sizeof (head
),
418 .module
= dbs
[cnt
].suggested_module
,
419 .data_size
= (dbs
[cnt
].suggested_module
420 * DEFAULT_DATASIZE_PER_BUCKET
),
425 if ((TEMP_FAILURE_RETRY (write (fd
, &head
, sizeof (head
)))
427 || ftruncate (fd
, total
) != 0
428 || (mem
= mmap (NULL
, total
, PROT_READ
| PROT_WRITE
,
429 MAP_SHARED
, fd
, 0)) == MAP_FAILED
)
431 unlink (dbs
[cnt
].db_filename
);
432 dbg_log (_("cannot write to database file %s: %s"),
433 dbs
[cnt
].db_filename
, strerror (errno
));
434 dbs
[cnt
].persistent
= 0;
440 dbs
[cnt
].data
= (char *)
441 &dbs
[cnt
].head
->array
[roundup (dbs
[cnt
].head
->module
,
442 ALIGN
/ sizeof (ref_t
))];
443 dbs
[cnt
].memsize
= total
;
444 dbs
[cnt
].mmap_used
= true;
446 /* Remember the descriptors. */
448 dbs
[cnt
].ro_fd
= ro_fd
;
461 && ((dbs
[cnt
].wr_fd
!= -1
462 && fcntl (dbs
[cnt
].wr_fd
, F_SETFD
, FD_CLOEXEC
) == -1)
463 || (dbs
[cnt
].ro_fd
!= -1
464 && fcntl (dbs
[cnt
].ro_fd
, F_SETFD
, FD_CLOEXEC
) == -1)))
467 cannot set socket to close on exec: %s; disabling paranoia mode"),
472 if (dbs
[cnt
].head
== NULL
)
474 /* We do not use the persistent database. Just
475 create an in-memory data structure. */
476 assert (! dbs
[cnt
].persistent
);
478 dbs
[cnt
].head
= xmalloc (sizeof (struct database_pers_head
)
479 + (dbs
[cnt
].suggested_module
481 memset (dbs
[cnt
].head
, '\0', sizeof (dbs
[cnt
].head
));
482 assert (~ENDREF
== 0);
483 memset (dbs
[cnt
].head
->array
, '\xff',
484 dbs
[cnt
].suggested_module
* sizeof (ref_t
));
485 dbs
[cnt
].head
->module
= dbs
[cnt
].suggested_module
;
486 dbs
[cnt
].head
->data_size
= (DEFAULT_DATASIZE_PER_BUCKET
487 * dbs
[cnt
].head
->module
);
488 dbs
[cnt
].data
= xmalloc (dbs
[cnt
].head
->data_size
);
489 dbs
[cnt
].head
->first_free
= 0;
492 assert (dbs
[cnt
].ro_fd
== -1);
495 if (dbs
[cnt
].check_file
)
497 /* We need the modification date of the file. */
500 if (stat (dbs
[cnt
].filename
, &st
) < 0)
502 /* We cannot stat() the file, disable file checking. */
503 dbg_log (_("cannot stat() file `%s': %s"),
504 dbs
[cnt
].filename
, strerror (errno
));
505 dbs
[cnt
].check_file
= 0;
508 dbs
[cnt
].file_mtime
= st
.st_mtime
;
512 /* Create the socket. */
513 sock
= socket (AF_UNIX
, SOCK_STREAM
, 0);
516 dbg_log (_("cannot open socket: %s"), strerror (errno
));
519 /* Bind a name to the socket. */
520 struct sockaddr_un sock_addr
;
521 sock_addr
.sun_family
= AF_UNIX
;
522 strcpy (sock_addr
.sun_path
, _PATH_NSCDSOCKET
);
523 if (bind (sock
, (struct sockaddr
*) &sock_addr
, sizeof (sock_addr
)) < 0)
525 dbg_log ("%s: %s", _PATH_NSCDSOCKET
, strerror (errno
));
529 /* We don't want to get stuck on accept. */
530 int fl
= fcntl (sock
, F_GETFL
);
531 if (fl
== -1 || fcntl (sock
, F_SETFL
, fl
| O_NONBLOCK
) == -1)
533 dbg_log (_("cannot change socket to nonblocking mode: %s"),
538 /* The descriptor needs to be closed on exec. */
539 if (paranoia
&& fcntl (sock
, F_SETFD
, FD_CLOEXEC
) == -1)
541 dbg_log (_("cannot set socket to close on exec: %s"),
546 /* Set permissions for the socket. */
547 chmod (_PATH_NSCDSOCKET
, DEFFILEMODE
);
549 /* Set the socket up to accept connections. */
550 if (listen (sock
, SOMAXCONN
) < 0)
552 dbg_log (_("cannot enable socket to accept connections: %s"),
557 /* Change to unprivileged uid/gid/groups if specifed in config file */
558 if (server_user
!= NULL
)
559 finish_drop_privileges ();
563 /* Close the connections. */
572 invalidate_cache (char *key
)
576 if (strcmp (key
, "passwd") == 0)
578 else if (strcmp (key
, "group") == 0)
580 else if (__builtin_expect (strcmp (key
, "hosts"), 0) == 0)
584 /* Re-initialize the resolver. resolv.conf might have changed. */
590 if (dbs
[number
].enabled
)
591 prune_cache (&dbs
[number
], LONG_MAX
);
597 send_ro_fd (struct database_dyn
*db
, char *key
, int fd
)
599 /* If we do not have an read-only file descriptor do nothing. */
603 /* We need to send some data along with the descriptor. */
605 iov
[0].iov_base
= key
;
606 iov
[0].iov_len
= strlen (key
) + 1;
608 /* Prepare the control message to transfer the descriptor. */
609 char buf
[CMSG_SPACE (sizeof (int))];
610 struct msghdr msg
= { .msg_iov
= iov
, .msg_iovlen
= 1,
611 .msg_control
= buf
, .msg_controllen
= sizeof (buf
) };
612 struct cmsghdr
*cmsg
= CMSG_FIRSTHDR (&msg
);
614 cmsg
->cmsg_level
= SOL_SOCKET
;
615 cmsg
->cmsg_type
= SCM_RIGHTS
;
616 cmsg
->cmsg_len
= CMSG_LEN (sizeof (int));
618 *(int *) CMSG_DATA (cmsg
) = db
->ro_fd
;
620 msg
.msg_controllen
= cmsg
->cmsg_len
;
622 /* Send the control message. We repeat when we are interrupted but
623 everything else is ignored. */
625 # define MSG_NOSIGNAL 0
627 (void) TEMP_FAILURE_RETRY (sendmsg (fd
, &msg
, MSG_NOSIGNAL
));
629 if (__builtin_expect (debug_level
> 0, 0))
630 dbg_log (_("provide access to FD %d, for %s"), db
->ro_fd
, key
);
632 #endif /* SCM_RIGHTS */
635 /* Handle new request. */
637 handle_request (int fd
, request_header
*req
, void *key
, uid_t uid
)
639 if (__builtin_expect (req
->version
, NSCD_VERSION
) != NSCD_VERSION
)
643 cannot handle old request version %d; current version is %d"),
644 req
->version
, NSCD_VERSION
);
648 /* Make the SELinux check before we go on to the standard checks. We
649 need to verify that the request type is valid, since it has not
650 yet been checked at this point. */
652 && __builtin_expect (req
->type
, GETPWBYNAME
) >= GETPWBYNAME
653 && __builtin_expect (req
->type
, LASTREQ
) < LASTREQ
654 && nscd_request_avc_has_perm (fd
, req
->type
) != 0)
657 struct database_dyn
*db
= serv2db
[req
->type
];
659 // XXX Clean up so that each new command need not introduce a
660 // XXX new conditional.
661 if ((__builtin_expect (req
->type
, GETPWBYNAME
) >= GETPWBYNAME
662 && __builtin_expect (req
->type
, LASTDBREQ
) <= LASTDBREQ
)
663 || req
->type
== GETAI
|| req
->type
== INITGROUPS
)
665 if (__builtin_expect (debug_level
, 0) > 0)
667 if (req
->type
== GETHOSTBYADDR
|| req
->type
== GETHOSTBYADDRv6
)
669 char buf
[INET6_ADDRSTRLEN
];
671 dbg_log ("\t%s (%s)", serv2str
[req
->type
],
672 inet_ntop (req
->type
== GETHOSTBYADDR
673 ? AF_INET
: AF_INET6
,
674 key
, buf
, sizeof (buf
)));
677 dbg_log ("\t%s (%s)", serv2str
[req
->type
], (char *) key
);
680 /* Is this service enabled? */
683 /* No, sent the prepared record. */
684 if (TEMP_FAILURE_RETRY (write (fd
, db
->disabled_iov
->iov_base
,
685 db
->disabled_iov
->iov_len
))
686 != (ssize_t
) db
->disabled_iov
->iov_len
687 && __builtin_expect (debug_level
, 0) > 0)
689 /* We have problems sending the result. */
691 dbg_log (_("cannot write result: %s"),
692 strerror_r (errno
, buf
, sizeof (buf
)));
698 /* Be sure we can read the data. */
699 if (__builtin_expect (pthread_rwlock_tryrdlock (&db
->lock
) != 0, 0))
701 ++db
->head
->rdlockdelayed
;
702 pthread_rwlock_rdlock (&db
->lock
);
705 /* See whether we can handle it from the cache. */
706 struct datahead
*cached
;
707 cached
= (struct datahead
*) cache_search (req
->type
, key
, req
->key_len
,
711 /* Hurray it's in the cache. */
712 if (writeall (fd
, cached
->data
, cached
->recsize
)
714 && __builtin_expect (debug_level
, 0) > 0)
716 /* We have problems sending the result. */
718 dbg_log (_("cannot write result: %s"),
719 strerror_r (errno
, buf
, sizeof (buf
)));
722 pthread_rwlock_unlock (&db
->lock
);
727 pthread_rwlock_unlock (&db
->lock
);
729 else if (__builtin_expect (debug_level
, 0) > 0)
731 if (req
->type
== INVALIDATE
)
732 dbg_log ("\t%s (%s)", serv2str
[req
->type
], (char *) key
);
734 dbg_log ("\t%s", serv2str
[req
->type
]);
737 /* Handle the request. */
741 addpwbyname (db
, fd
, req
, key
, uid
);
745 addpwbyuid (db
, fd
, req
, key
, uid
);
749 addgrbyname (db
, fd
, req
, key
, uid
);
753 addgrbygid (db
, fd
, req
, key
, uid
);
757 addhstbyname (db
, fd
, req
, key
, uid
);
760 case GETHOSTBYNAMEv6
:
761 addhstbynamev6 (db
, fd
, req
, key
, uid
);
765 addhstbyaddr (db
, fd
, req
, key
, uid
);
768 case GETHOSTBYADDRv6
:
769 addhstbyaddrv6 (db
, fd
, req
, key
, uid
);
773 addhstai (db
, fd
, req
, key
, uid
);
777 addinitgroups (db
, fd
, req
, key
, uid
);
785 /* Get the callers credentials. */
788 socklen_t optlen
= sizeof (caller
);
790 if (getsockopt (fd
, SOL_SOCKET
, SO_PEERCRED
, &caller
, &optlen
) < 0)
794 dbg_log (_("error getting callers id: %s"),
795 strerror_r (errno
, buf
, sizeof (buf
)));
801 /* Some systems have no SO_PEERCRED implementation. They don't
802 care about security so we don't as well. */
807 /* Accept shutdown, getstat and invalidate only from root. For
808 the stat call also allow the user specified in the config file. */
809 if (req
->type
== GETSTAT
)
811 if (uid
== 0 || uid
== stat_uid
)
812 send_stats (fd
, dbs
);
816 if (req
->type
== INVALIDATE
)
817 invalidate_cache (key
);
819 termination_handler (0);
827 send_ro_fd (serv2db
[req
->type
], key
, fd
);
832 /* Ignore the command, it's nothing we know. */
838 /* Restart the process. */
842 /* First determine the parameters. We do not use the parameters
843 passed to main() since in case nscd is started by running the
844 dynamic linker this will not work. Yes, this is not the usual
845 case but nscd is part of glibc and we occasionally do this. */
846 size_t buflen
= 1024;
847 char *buf
= alloca (buflen
);
849 int fd
= open ("/proc/self/cmdline", O_RDONLY
);
853 cannot open /proc/self/cmdline: %s; disabling paranoia mode"),
862 ssize_t n
= TEMP_FAILURE_RETRY (read (fd
, buf
+ readlen
,
867 cannot open /proc/self/cmdline: %s; disabling paranoia mode"),
877 if (readlen
< buflen
)
880 /* We might have to extend the buffer. */
881 size_t old_buflen
= buflen
;
882 char *newp
= extend_alloca (buf
, buflen
, 2 * buflen
);
883 buf
= memmove (newp
, buf
, old_buflen
);
888 /* Parse the command line. Worst case scenario: every two
889 characters form one parameter (one character plus NUL). */
890 char **argv
= alloca ((readlen
/ 2 + 1) * sizeof (argv
[0]));
894 while (cp
< buf
+ readlen
)
897 cp
= (char *) rawmemchr (cp
, '\0') + 1;
901 /* Second, change back to the old user if we changed it. */
902 if (server_user
!= NULL
)
904 if (setuid (old_uid
) != 0)
907 cannot change to old UID: %s; disabling paranoia mode"),
914 if (setgid (old_gid
) != 0)
917 cannot change to old GID: %s; disabling paranoia mode"),
926 /* Next change back to the old working directory. */
927 if (chdir (oldcwd
) == -1)
930 cannot change to old working directory: %s; disabling paranoia mode"),
933 if (server_user
!= NULL
)
942 /* Synchronize memory. */
943 for (int cnt
= 0; cnt
< lastdb
; ++cnt
)
945 /* Make sure nobody keeps using the database. */
946 dbs
[cnt
].head
->timestamp
= 0;
948 if (dbs
[cnt
].persistent
)
950 msync (dbs
[cnt
].head
, dbs
[cnt
].memsize
, MS_ASYNC
);
953 /* The preparations are done. */
954 execv ("/proc/self/exe", argv
);
956 /* If we come here, we will never be able to re-exec. */
957 dbg_log (_("re-exec failed: %s; disabling paranoia mode"),
960 if (server_user
!= NULL
)
970 /* List of file descriptors. */
976 /* Memory allocated for the list. */
977 static struct fdlist
*fdlist
;
978 /* List of currently ready-to-read file descriptors. */
979 static struct fdlist
*readylist
;
981 /* Conditional variable and mutex to signal availability of entries in
982 READYLIST. The condvar is initialized dynamically since we might
983 use a different clock depending on availability. */
984 static pthread_cond_t readylist_cond
;
985 static pthread_mutex_t readylist_lock
= PTHREAD_MUTEX_INITIALIZER
;
987 /* The clock to use with the condvar. */
988 static clockid_t timeout_clock
= CLOCK_REALTIME
;
990 /* Number of threads ready to handle the READYLIST. */
991 static unsigned long int nready
;
994 /* This is the main loop. It is replicated in different threads but the
995 `poll' call makes sure only one thread handles an incoming connection. */
997 __attribute__ ((__noreturn__
))
1000 const long int my_number
= (long int) p
;
1001 const int run_prune
= my_number
< lastdb
&& dbs
[my_number
].enabled
;
1002 struct timespec prune_ts
;
1008 setup_thread (&dbs
[my_number
]);
1010 /* We are running. */
1011 dbs
[my_number
].head
->timestamp
= time (NULL
);
1013 if (clock_gettime (timeout_clock
, &prune_ts
) == -1)
1014 /* Should never happen. */
1017 /* Compute timeout time. */
1018 prune_ts
.tv_sec
+= CACHE_PRUNE_INTERVAL
;
1021 /* Initial locking. */
1022 pthread_mutex_lock (&readylist_lock
);
1024 /* One more thread available. */
1029 while (readylist
== NULL
)
1033 /* Wait, but not forever. */
1034 to
= pthread_cond_timedwait (&readylist_cond
, &readylist_lock
,
1037 /* If we were woken and there is no work to be done,
1038 just start pruning. */
1039 if (readylist
== NULL
&& to
== ETIMEDOUT
)
1042 pthread_mutex_unlock (&readylist_lock
);
1047 /* No need to timeout. */
1048 pthread_cond_wait (&readylist_cond
, &readylist_lock
);
1051 struct fdlist
*it
= readylist
->next
;
1052 if (readylist
->next
== readylist
)
1053 /* Just one entry on the list. */
1056 readylist
->next
= it
->next
;
1058 /* Extract the information and mark the record ready to be used
1063 /* One more thread available. */
1066 /* We are done with the list. */
1067 pthread_mutex_unlock (&readylist_lock
);
1069 /* We do not want to block on a short read or so. */
1070 int fl
= fcntl (fd
, F_GETFL
);
1071 if (fl
== -1 || fcntl (fd
, F_SETFL
, fl
| O_NONBLOCK
) == -1)
1074 /* Now read the request. */
1076 if (__builtin_expect (TEMP_FAILURE_RETRY (read (fd
, &req
, sizeof (req
)))
1077 != sizeof (req
), 0))
1079 /* We failed to read data. Note that this also might mean we
1080 failed because we would have blocked. */
1081 if (debug_level
> 0)
1082 dbg_log (_("short read while reading request: %s"),
1083 strerror_r (errno
, buf
, sizeof (buf
)));
1087 /* Check whether this is a valid request type. */
1088 if (req
.type
< GETPWBYNAME
|| req
.type
>= LASTREQ
)
1091 /* Some systems have no SO_PEERCRED implementation. They don't
1092 care about security so we don't as well. */
1099 struct ucred caller
;
1100 socklen_t optlen
= sizeof (caller
);
1102 if (getsockopt (fd
, SOL_SOCKET
, SO_PEERCRED
, &caller
, &optlen
) < 0)
1104 dbg_log (_("error getting callers id: %s"),
1105 strerror_r (errno
, buf
, sizeof (buf
)));
1109 if (req
.type
< GETPWBYNAME
|| req
.type
> LASTDBREQ
1110 || serv2db
[req
.type
]->secure
)
1115 else if (__builtin_expect (debug_level
> 0, 0))
1117 struct ucred caller
;
1118 socklen_t optlen
= sizeof (caller
);
1120 if (getsockopt (fd
, SOL_SOCKET
, SO_PEERCRED
, &caller
, &optlen
) == 0)
1125 /* It should not be possible to crash the nscd with a silly
1126 request (i.e., a terribly large key). We limit the size to 1kb. */
1127 #define MAXKEYLEN 1024
1128 if (__builtin_expect (req
.key_len
, 1) < 0
1129 || __builtin_expect (req
.key_len
, 1) > MAXKEYLEN
)
1131 if (debug_level
> 0)
1132 dbg_log (_("key length in request too long: %d"), req
.key_len
);
1137 char keybuf
[MAXKEYLEN
];
1139 if (__builtin_expect (TEMP_FAILURE_RETRY (read (fd
, keybuf
,
1143 /* Again, this can also mean we would have blocked. */
1144 if (debug_level
> 0)
1145 dbg_log (_("short read while reading request key: %s"),
1146 strerror_r (errno
, buf
, sizeof (buf
)));
1150 if (__builtin_expect (debug_level
, 0) > 0)
1155 handle_request: request received (Version = %d) from PID %ld"),
1156 req
.version
, (long int) pid
);
1160 handle_request: request received (Version = %d)"), req
.version
);
1163 /* Phew, we got all the data, now process it. */
1164 handle_request (fd
, &req
, keybuf
, uid
);
1171 /* Check whether we should be pruning the cache. */
1172 assert (run_prune
|| to
== 0);
1173 if (to
== ETIMEDOUT
)
1176 /* The pthread_cond_timedwait() call timed out. It is time
1177 to clean up the cache. */
1178 assert (my_number
< lastdb
);
1179 prune_cache (&dbs
[my_number
],
1180 prune_ts
.tv_sec
+ (prune_ts
.tv_nsec
>= 500000000));
1182 if (clock_gettime (timeout_clock
, &prune_ts
) == -1)
1183 /* Should never happen. */
1186 /* Compute next timeout time. */
1187 prune_ts
.tv_sec
+= CACHE_PRUNE_INTERVAL
;
1189 /* In case the list is emtpy we do not want to run the prune
1190 code right away again. */
1195 pthread_mutex_lock (&readylist_lock
);
1197 /* One more thread available. */
1203 static unsigned int nconns
;
1208 pthread_mutex_lock (&readylist_lock
);
1210 /* Find an empty entry in FDLIST. */
1212 for (inner
= 0; inner
< nconns
; ++inner
)
1213 if (fdlist
[inner
].next
== NULL
)
1215 assert (inner
< nconns
);
1217 fdlist
[inner
].fd
= fd
;
1219 if (readylist
== NULL
)
1220 readylist
= fdlist
[inner
].next
= &fdlist
[inner
];
1223 fdlist
[inner
].next
= readylist
->next
;
1224 readylist
= readylist
->next
= &fdlist
[inner
];
1227 bool do_signal
= true;
1228 if (__builtin_expect (nready
== 0, 0))
1233 /* Try to start another thread to help out. */
1235 if (nthreads
< max_nthreads
1236 && pthread_create (&th
, &attr
, nscd_run
,
1237 (void *) (long int) nthreads
) == 0)
1239 /* We got another thread. */
1241 /* The new thread might new a kick. */
1247 pthread_mutex_unlock (&readylist_lock
);
1249 /* Tell one of the worker threads there is work to do. */
1251 pthread_cond_signal (&readylist_cond
);
1255 /* Check whether restarting should happen. */
1257 restart_p (time_t now
)
1259 return (paranoia
&& readylist
== NULL
&& nready
== nthreads
1260 && now
>= restart_time
);
1264 /* Array for times a connection was accepted. */
1265 static time_t *starttime
;
1269 __attribute__ ((__noreturn__
))
1270 main_loop_poll (void)
1272 struct pollfd
*conns
= (struct pollfd
*) xmalloc (nconns
1273 * sizeof (conns
[0]));
1276 conns
[0].events
= POLLRDNORM
;
1278 size_t firstfree
= 1;
1282 /* Wait for any event. We wait at most a couple of seconds so
1283 that we can check whether we should close any of the accepted
1284 connections since we have not received a request. */
1285 #define MAX_ACCEPT_TIMEOUT 30
1286 #define MIN_ACCEPT_TIMEOUT 5
1287 #define MAIN_THREAD_TIMEOUT \
1288 (MAX_ACCEPT_TIMEOUT * 1000 \
1289 - ((MAX_ACCEPT_TIMEOUT - MIN_ACCEPT_TIMEOUT) * 1000 * nused) / (2 * nconns))
1291 int n
= poll (conns
, nused
, MAIN_THREAD_TIMEOUT
);
1293 time_t now
= time (NULL
);
1295 /* If there is a descriptor ready for reading or there is a new
1296 connection, process this now. */
1299 if (conns
[0].revents
!= 0)
1301 /* We have a new incoming connection. Accept the connection. */
1302 int fd
= TEMP_FAILURE_RETRY (accept (sock
, NULL
, NULL
));
1304 /* use the descriptor if we have not reached the limit. */
1305 if (fd
>= 0 && firstfree
< nconns
)
1307 conns
[firstfree
].fd
= fd
;
1308 conns
[firstfree
].events
= POLLRDNORM
;
1309 starttime
[firstfree
] = now
;
1310 if (firstfree
>= nused
)
1311 nused
= firstfree
+ 1;
1315 while (firstfree
< nused
&& conns
[firstfree
].fd
!= -1);
1321 for (size_t cnt
= 1; cnt
< nused
&& n
> 0; ++cnt
)
1322 if (conns
[cnt
].revents
!= 0)
1324 fd_ready (conns
[cnt
].fd
);
1326 /* Clean up the CONNS array. */
1328 if (cnt
< firstfree
)
1330 if (cnt
== nused
- 1)
1333 while (conns
[nused
- 1].fd
== -1);
1339 /* Now find entries which have timed out. */
1342 /* We make the timeout length depend on the number of file
1343 descriptors currently used. */
1344 #define ACCEPT_TIMEOUT \
1345 (MAX_ACCEPT_TIMEOUT \
1346 - ((MAX_ACCEPT_TIMEOUT - MIN_ACCEPT_TIMEOUT) * nused) / nconns)
1347 time_t laststart
= now
- ACCEPT_TIMEOUT
;
1349 for (size_t cnt
= nused
- 1; cnt
> 0; --cnt
)
1351 if (conns
[cnt
].fd
!= -1 && starttime
[cnt
] < laststart
)
1353 /* Remove the entry, it timed out. */
1354 (void) close (conns
[cnt
].fd
);
1357 if (cnt
< firstfree
)
1359 if (cnt
== nused
- 1)
1362 while (conns
[nused
- 1].fd
== -1);
1366 if (restart_p (now
))
1374 main_loop_epoll (int efd
)
1376 struct epoll_event ev
= { 0, };
1380 /* Add the socket. */
1381 ev
.events
= EPOLLRDNORM
;
1383 if (epoll_ctl (efd
, EPOLL_CTL_ADD
, sock
, &ev
) == -1)
1384 /* We cannot use epoll. */
1389 struct epoll_event revs
[100];
1390 # define nrevs (sizeof (revs) / sizeof (revs[0]))
1392 int n
= epoll_wait (efd
, revs
, nrevs
, MAIN_THREAD_TIMEOUT
);
1394 time_t now
= time (NULL
);
1396 for (int cnt
= 0; cnt
< n
; ++cnt
)
1397 if (revs
[cnt
].data
.fd
== sock
)
1399 /* A new connection. */
1400 int fd
= TEMP_FAILURE_RETRY (accept (sock
, NULL
, NULL
));
1404 /* Try to add the new descriptor. */
1407 || epoll_ctl (efd
, EPOLL_CTL_ADD
, fd
, &ev
) == -1)
1408 /* The descriptor is too large or something went
1409 wrong. Close the descriptor. */
1413 /* Remember when we accepted the connection. */
1414 starttime
[fd
] = now
;
1425 /* Remove the descriptor from the epoll descriptor. */
1426 struct epoll_event ev
= { 0, };
1427 (void) epoll_ctl (efd
, EPOLL_CTL_DEL
, revs
[cnt
].data
.fd
, &ev
);
1429 /* Get a worked to handle the request. */
1430 fd_ready (revs
[cnt
].data
.fd
);
1432 /* Reset the time. */
1433 starttime
[revs
[cnt
].data
.fd
] = 0;
1434 if (revs
[cnt
].data
.fd
== highest
)
1437 while (highest
> 0 && starttime
[highest
] == 0);
1442 /* Now look for descriptors for accepted connections which have
1443 no reply in too long of a time. */
1444 time_t laststart
= now
- ACCEPT_TIMEOUT
;
1445 for (int cnt
= highest
; cnt
> STDERR_FILENO
; --cnt
)
1446 if (cnt
!= sock
&& starttime
[cnt
] != 0 && starttime
[cnt
] < laststart
)
1448 /* We are waiting for this one for too long. Close it. */
1449 struct epoll_event ev
= {0, };
1450 (void) epoll_ctl (efd
, EPOLL_CTL_DEL
, cnt
, &ev
);
1458 else if (cnt
!= sock
&& starttime
[cnt
] == 0 && cnt
== highest
)
1461 if (restart_p (now
))
1468 /* Start all the threads we want. The initial process is thread no. 1. */
1470 start_threads (void)
1472 /* Initialize the conditional variable we will use. The only
1473 non-standard attribute we might use is the clock selection. */
1474 pthread_condattr_t condattr
;
1475 pthread_condattr_init (&condattr
);
1477 #if defined _POSIX_MONOTONIC_CLOCK && _POSIX_MONOTONIC_CLOCK >= 0
1478 /* Determine whether the monotonous clock is available. */
1479 struct timespec dummy
;
1480 # if _POSIX_MONOTONIC_CLOCK == 0
1481 if (sysconf (_SC_MONOTONIC_CLOCK
) > 0)
1484 # if defined _POSIX_CLOCK_SELECTION && _POSIX_CLOCK_SELECTION >= 0
1485 # if _POSIX_CLOCK_SELECTION == 0
1486 if (sysconf (_SC_CLOCK_SELECTION
) > 0)
1488 if (clock_getres (CLOCK_MONOTONIC
, &dummy
) == 0
1489 && pthread_condattr_setclock (&condattr
, CLOCK_MONOTONIC
) == 0)
1490 timeout_clock
= CLOCK_MONOTONIC
;
1491 # elif _POSIX_THREADS > 0
1492 if (sysconf (_SC_CLOCK_SELECTION
) > 0)
1494 void *h
= __libc_dlopen (LIBPTHREAD_SO
);
1495 int (*condattr_setclock
) (pthread_condattr_t
*, __clockid_t
) = NULL
;
1498 condattr_setclock
= __libc_dlsym (h
, "pthread_condattr_setclock");
1500 if (condattr_setclock
1501 && clock_getres (CLOCK_MONOTONIC
, &dummy
) == 0
1502 && condattr_setclock (&condattr
, CLOCK_MONOTONIC
) == 0)
1503 timeout_clock
= CLOCK_MONOTONIC
;
1512 pthread_cond_init (&readylist_cond
, &condattr
);
1513 pthread_condattr_destroy (&condattr
);
1516 /* Create the attribute for the threads. They are all created
1518 pthread_attr_init (&attr
);
1519 pthread_attr_setdetachstate (&attr
, PTHREAD_CREATE_DETACHED
);
1520 /* Use 1MB stacks, twice as much for 64-bit architectures. */
1521 pthread_attr_setstacksize (&attr
, 1024 * 1024 * (sizeof (void *) / 4));
1523 /* We allow less than LASTDB threads only for debugging. */
1524 if (debug_level
== 0)
1525 nthreads
= MAX (nthreads
, lastdb
);
1528 for (long int i
= 0; i
< nthreads
; ++i
)
1531 if (pthread_create (&th
, &attr
, nscd_run
, (void *) (i
- nfailed
)) != 0)
1534 if (nthreads
- nfailed
< lastdb
)
1536 /* We could not start enough threads. */
1537 dbg_log (_("could only start %d threads; terminating"),
1538 nthreads
- nfailed
);
1542 /* Determine how much room for descriptors we should initially
1543 allocate. This might need to change later if we cap the number
1545 const long int nfds
= sysconf (_SC_OPEN_MAX
);
1547 #define MAXCONN 16384
1548 if (nfds
== -1 || nfds
> MAXCONN
)
1550 else if (nfds
< MINCONN
)
1555 /* We need memory to pass descriptors on to the worker threads. */
1556 fdlist
= (struct fdlist
*) xcalloc (nconns
, sizeof (fdlist
[0]));
1557 /* Array to keep track when connection was accepted. */
1558 starttime
= (time_t *) xcalloc (nconns
, sizeof (starttime
[0]));
1560 /* In the main thread we execute the loop which handles incoming
1563 int efd
= epoll_create (100);
1566 main_loop_epoll (efd
);
1574 /* Look up the uid, gid, and supplementary groups to run nscd as. When
1575 this function is called, we are not listening on the nscd socket yet so
1576 we can just use the ordinary lookup functions without causing a lockup */
1578 begin_drop_privileges (void)
1580 struct passwd
*pwd
= getpwnam (server_user
);
1584 dbg_log (_("Failed to run nscd as user '%s'"), server_user
);
1585 error (EXIT_FAILURE
, 0, _("Failed to run nscd as user '%s'"),
1589 server_uid
= pwd
->pw_uid
;
1590 server_gid
= pwd
->pw_gid
;
1592 /* Save the old UID/GID if we have to change back. */
1595 old_uid
= getuid ();
1596 old_gid
= getgid ();
1599 if (getgrouplist (server_user
, server_gid
, NULL
, &server_ngroups
) == 0)
1601 /* This really must never happen. */
1602 dbg_log (_("Failed to run nscd as user '%s'"), server_user
);
1603 error (EXIT_FAILURE
, errno
, _("initial getgrouplist failed"));
1606 server_groups
= (gid_t
*) xmalloc (server_ngroups
* sizeof (gid_t
));
1608 if (getgrouplist (server_user
, server_gid
, server_groups
, &server_ngroups
)
1611 dbg_log (_("Failed to run nscd as user '%s'"), server_user
);
1612 error (EXIT_FAILURE
, errno
, _("getgrouplist failed"));
1617 /* Call setgroups(), setgid(), and setuid() to drop root privileges and
1618 run nscd as the user specified in the configuration file. */
1620 finish_drop_privileges (void)
1622 if (setgroups (server_ngroups
, server_groups
) == -1)
1624 dbg_log (_("Failed to run nscd as user '%s'"), server_user
);
1625 error (EXIT_FAILURE
, errno
, _("setgroups failed"));
1628 if (setgid (server_gid
) == -1)
1630 dbg_log (_("Failed to run nscd as user '%s'"), server_user
);
1635 if (setuid (server_uid
) == -1)
1637 dbg_log (_("Failed to run nscd as user '%s'"), server_user
);
1643 /* Handle the HUP signal which will force a dump of the cache */
1645 sighup_handler (int signum
)
1647 /* Prune the password database */
1648 prune_cache (&dbs
[pwddb
], LONG_MAX
);
1650 /* Prune the group database */
1651 prune_cache (&dbs
[grpdb
], LONG_MAX
);
1653 /* Prune the host database */
1654 prune_cache (&dbs
[hstdb
], LONG_MAX
);