]>
Commit | Line | Data |
---|---|---|
67479a70 | 1 | /* Inner loops of cache daemon. |
43397eaf | 2 | Copyright (C) 1998-2003, 2004, 2005, 2006, 2007 Free Software Foundation, Inc. |
d67281a7 | 3 | This file is part of the GNU C Library. |
67479a70 | 4 | Contributed by Ulrich Drepper <drepper@cygnus.com>, 1998. |
d67281a7 | 5 | |
43bc8ac6 UD |
6 | This program is free software; you can redistribute it and/or modify |
7 | it under the terms of the GNU General Public License version 2 as | |
8 | published by the Free Software Foundation. | |
d67281a7 | 9 | |
43bc8ac6 | 10 | This program is distributed in the hope that it will be useful, |
d67281a7 | 11 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
43bc8ac6 UD |
12 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
13 | GNU General Public License for more details. | |
d67281a7 | 14 | |
43bc8ac6 UD |
15 | You should have received a copy of the GNU General Public License |
16 | along with this program; if not, write to the Free Software Foundation, | |
17 | Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */ | |
d67281a7 | 18 | |
4401d759 | 19 | #include <alloca.h> |
67479a70 | 20 | #include <assert.h> |
0fdb4f42 | 21 | #include <atomic.h> |
d67281a7 | 22 | #include <error.h> |
67479a70 | 23 | #include <errno.h> |
d6db0975 | 24 | #include <fcntl.h> |
057685e4 | 25 | #include <grp.h> |
a95a08b4 | 26 | #include <libintl.h> |
d67281a7 | 27 | #include <pthread.h> |
057685e4 | 28 | #include <pwd.h> |
482bbeb9 | 29 | #include <resolv.h> |
057685e4 | 30 | #include <stdio.h> |
d67281a7 UD |
31 | #include <stdlib.h> |
32 | #include <unistd.h> | |
8d8c6efa | 33 | #include <arpa/inet.h> |
fc03df7a UD |
34 | #ifdef HAVE_EPOLL |
35 | # include <sys/epoll.h> | |
36 | #endif | |
a95a08b4 | 37 | #include <sys/mman.h> |
67479a70 | 38 | #include <sys/param.h> |
a53bad16 | 39 | #include <sys/poll.h> |
eac10791 UD |
40 | #ifdef HAVE_SENDFILE |
41 | # include <sys/sendfile.h> | |
42 | #endif | |
d67281a7 UD |
43 | #include <sys/socket.h> |
44 | #include <sys/stat.h> | |
d67281a7 UD |
45 | #include <sys/un.h> |
46 | ||
47 | #include "nscd.h" | |
48 | #include "dbg_log.h" | |
74a30a58 | 49 | #include "selinux.h" |
eac10791 UD |
50 | #ifdef HAVE_SENDFILE |
51 | # include <kernel-features.h> | |
52 | #endif | |
a334319f UD |
53 | |
54 | ||
057685e4 UD |
55 | /* Wrapper functions with error checking for standard functions. */ |
56 | extern void *xmalloc (size_t n); | |
57 | extern void *xcalloc (size_t n, size_t s); | |
58 | extern void *xrealloc (void *o, size_t n); | |
59 | ||
60 | /* Support to run nscd as an unprivileged user */ | |
61 | const char *server_user; | |
62 | static uid_t server_uid; | |
63 | static gid_t server_gid; | |
a12ce44f UD |
64 | const char *stat_user; |
65 | uid_t stat_uid; | |
057685e4 UD |
66 | static gid_t *server_groups; |
67 | #ifndef NGROUPS | |
68 | # define NGROUPS 32 | |
69 | #endif | |
a95a08b4 | 70 | static int server_ngroups; |
057685e4 | 71 | |
27e82856 UD |
72 | static pthread_attr_t attr; |
73 | ||
057685e4 UD |
74 | static void begin_drop_privileges (void); |
75 | static void finish_drop_privileges (void); | |
76 | ||
67479a70 | 77 | /* Map request type to a string. */ |
b21fa963 | 78 | const char *const serv2str[LASTREQ] = |
d67281a7 | 79 | { |
67479a70 UD |
80 | [GETPWBYNAME] = "GETPWBYNAME", |
81 | [GETPWBYUID] = "GETPWBYUID", | |
82 | [GETGRBYNAME] = "GETGRBYNAME", | |
83 | [GETGRBYGID] = "GETGRBYGID", | |
84 | [GETHOSTBYNAME] = "GETHOSTBYNAME", | |
85 | [GETHOSTBYNAMEv6] = "GETHOSTBYNAMEv6", | |
86 | [GETHOSTBYADDR] = "GETHOSTBYADDR", | |
87 | [GETHOSTBYADDRv6] = "GETHOSTBYADDRv6", | |
88 | [SHUTDOWN] = "SHUTDOWN", | |
756409c4 | 89 | [GETSTAT] = "GETSTAT", |
c207f23b UD |
90 | [INVALIDATE] = "INVALIDATE", |
91 | [GETFDPW] = "GETFDPW", | |
92 | [GETFDGR] = "GETFDGR", | |
d19687d6 | 93 | [GETFDHST] = "GETFDHST", |
f7e7a396 | 94 | [GETAI] = "GETAI", |
b21fa963 UD |
95 | [INITGROUPS] = "INITGROUPS", |
96 | [GETSERVBYNAME] = "GETSERVBYNAME", | |
97 | [GETSERVBYPORT] = "GETSERVBYPORT", | |
98 | [GETFDSERV] = "GETFDSERV" | |
67479a70 UD |
99 | }; |
100 | ||
101 | /* The control data structures for the services. */ | |
a95a08b4 | 102 | struct database_dyn dbs[lastdb] = |
67479a70 UD |
103 | { |
104 | [pwddb] = { | |
c2e13112 | 105 | .lock = PTHREAD_RWLOCK_WRITER_NONRECURSIVE_INITIALIZER_NP, |
cabfcde1 | 106 | .prunelock = PTHREAD_MUTEX_INITIALIZER, |
c2e13112 RM |
107 | .enabled = 0, |
108 | .check_file = 1, | |
a95a08b4 | 109 | .persistent = 0, |
797ed6f7 | 110 | .propagate = 1, |
c207f23b | 111 | .shared = 0, |
2c210d1e | 112 | .max_db_size = DEFAULT_MAX_DB_SIZE, |
b21fa963 | 113 | .reset_res = 0, |
c2e13112 | 114 | .filename = "/etc/passwd", |
a95a08b4 | 115 | .db_filename = _PATH_NSCD_PASSWD_DB, |
c2e13112 RM |
116 | .disabled_iov = &pwd_iov_disabled, |
117 | .postimeout = 3600, | |
a95a08b4 UD |
118 | .negtimeout = 20, |
119 | .wr_fd = -1, | |
120 | .ro_fd = -1, | |
121 | .mmap_used = false | |
67479a70 UD |
122 | }, |
123 | [grpdb] = { | |
c2e13112 | 124 | .lock = PTHREAD_RWLOCK_WRITER_NONRECURSIVE_INITIALIZER_NP, |
cabfcde1 | 125 | .prunelock = PTHREAD_MUTEX_INITIALIZER, |
c2e13112 RM |
126 | .enabled = 0, |
127 | .check_file = 1, | |
a95a08b4 | 128 | .persistent = 0, |
797ed6f7 | 129 | .propagate = 1, |
c207f23b | 130 | .shared = 0, |
2c210d1e | 131 | .max_db_size = DEFAULT_MAX_DB_SIZE, |
b21fa963 | 132 | .reset_res = 0, |
c2e13112 | 133 | .filename = "/etc/group", |
a95a08b4 | 134 | .db_filename = _PATH_NSCD_GROUP_DB, |
c2e13112 RM |
135 | .disabled_iov = &grp_iov_disabled, |
136 | .postimeout = 3600, | |
a95a08b4 UD |
137 | .negtimeout = 60, |
138 | .wr_fd = -1, | |
139 | .ro_fd = -1, | |
140 | .mmap_used = false | |
67479a70 UD |
141 | }, |
142 | [hstdb] = { | |
c2e13112 | 143 | .lock = PTHREAD_RWLOCK_WRITER_NONRECURSIVE_INITIALIZER_NP, |
cabfcde1 | 144 | .prunelock = PTHREAD_MUTEX_INITIALIZER, |
c2e13112 RM |
145 | .enabled = 0, |
146 | .check_file = 1, | |
a95a08b4 | 147 | .persistent = 0, |
797ed6f7 | 148 | .propagate = 0, /* Not used. */ |
c207f23b | 149 | .shared = 0, |
2c210d1e | 150 | .max_db_size = DEFAULT_MAX_DB_SIZE, |
b21fa963 | 151 | .reset_res = 1, |
c2e13112 | 152 | .filename = "/etc/hosts", |
a95a08b4 | 153 | .db_filename = _PATH_NSCD_HOSTS_DB, |
c2e13112 RM |
154 | .disabled_iov = &hst_iov_disabled, |
155 | .postimeout = 3600, | |
a95a08b4 UD |
156 | .negtimeout = 20, |
157 | .wr_fd = -1, | |
158 | .ro_fd = -1, | |
159 | .mmap_used = false | |
b21fa963 UD |
160 | }, |
161 | [servdb] = { | |
162 | .lock = PTHREAD_RWLOCK_WRITER_NONRECURSIVE_INITIALIZER_NP, | |
163 | .prunelock = PTHREAD_MUTEX_INITIALIZER, | |
164 | .enabled = 0, | |
165 | .check_file = 1, | |
166 | .persistent = 0, | |
167 | .propagate = 0, /* Not used. */ | |
168 | .shared = 0, | |
169 | .max_db_size = DEFAULT_MAX_DB_SIZE, | |
170 | .reset_res = 0, | |
171 | .filename = "/etc/services", | |
172 | .db_filename = _PATH_NSCD_SERVICES_DB, | |
173 | .disabled_iov = &serv_iov_disabled, | |
174 | .postimeout = 28800, | |
175 | .negtimeout = 20, | |
176 | .wr_fd = -1, | |
177 | .ro_fd = -1, | |
178 | .mmap_used = false | |
67479a70 UD |
179 | } |
180 | }; | |
d67281a7 | 181 | |
a95a08b4 UD |
182 | |
183 | /* Mapping of request type to database. */ | |
c207f23b | 184 | static struct database_dyn *const serv2db[LASTREQ] = |
a95a08b4 UD |
185 | { |
186 | [GETPWBYNAME] = &dbs[pwddb], | |
187 | [GETPWBYUID] = &dbs[pwddb], | |
188 | [GETGRBYNAME] = &dbs[grpdb], | |
189 | [GETGRBYGID] = &dbs[grpdb], | |
190 | [GETHOSTBYNAME] = &dbs[hstdb], | |
191 | [GETHOSTBYNAMEv6] = &dbs[hstdb], | |
192 | [GETHOSTBYADDR] = &dbs[hstdb], | |
c207f23b UD |
193 | [GETHOSTBYADDRv6] = &dbs[hstdb], |
194 | [GETFDPW] = &dbs[pwddb], | |
195 | [GETFDGR] = &dbs[grpdb], | |
196 | [GETFDHST] = &dbs[hstdb], | |
d19687d6 | 197 | [GETAI] = &dbs[hstdb], |
b21fa963 UD |
198 | [INITGROUPS] = &dbs[grpdb], |
199 | [GETSERVBYNAME] = &dbs[servdb], | |
200 | [GETSERVBYPORT] = &dbs[servdb], | |
201 | [GETFDSERV] = &dbs[servdb] | |
a95a08b4 UD |
202 | }; |
203 | ||
204 | ||
34489d95 UD |
205 | /* Number of seconds between two cache pruning runs. */ |
206 | #define CACHE_PRUNE_INTERVAL 15 | |
207 | ||
a95a08b4 | 208 | |
27e82856 | 209 | /* Initial number of threads to use. */ |
67479a70 | 210 | int nthreads = -1; |
27e82856 UD |
211 | /* Maximum number of threads to use. */ |
212 | int max_nthreads = 32; | |
d67281a7 | 213 | |
67479a70 UD |
214 | /* Socket for incoming connections. */ |
215 | static int sock; | |
d67281a7 | 216 | |
0fdb4f42 UD |
217 | /* Number of times clients had to wait. */ |
218 | unsigned long int client_queued; | |
219 | ||
d67281a7 | 220 | |
d2dc7d84 UD |
221 | ssize_t |
222 | writeall (int fd, const void *buf, size_t len) | |
223 | { | |
224 | size_t n = len; | |
225 | ssize_t ret; | |
226 | do | |
227 | { | |
2c210d1e | 228 | ret = TEMP_FAILURE_RETRY (send (fd, buf, n, MSG_NOSIGNAL)); |
d2dc7d84 UD |
229 | if (ret <= 0) |
230 | break; | |
231 | buf = (const char *) buf + ret; | |
232 | n -= ret; | |
233 | } | |
234 | while (n > 0); | |
235 | return ret < 0 ? ret : len - n; | |
236 | } | |
237 | ||
238 | ||
bd547139 UD |
239 | #ifdef HAVE_SENDFILE |
240 | ssize_t | |
241 | sendfileall (int tofd, int fromfd, off_t off, size_t len) | |
242 | { | |
243 | ssize_t n = len; | |
244 | ssize_t ret; | |
245 | ||
246 | do | |
247 | { | |
248 | ret = TEMP_FAILURE_RETRY (sendfile (tofd, fromfd, &off, n)); | |
249 | if (ret <= 0) | |
250 | break; | |
251 | n -= ret; | |
252 | } | |
253 | while (n > 0); | |
254 | return ret < 0 ? ret : len - n; | |
255 | } | |
256 | #endif | |
257 | ||
258 | ||
dc4bb1c2 UD |
259 | enum usekey |
260 | { | |
261 | use_not = 0, | |
262 | /* The following three are not really used, they are symbolic constants. */ | |
263 | use_first = 16, | |
264 | use_begin = 32, | |
265 | use_end = 64, | |
266 | ||
267 | use_he = 1, | |
268 | use_he_begin = use_he | use_begin, | |
269 | use_he_end = use_he | use_end, | |
270 | #if SEPARATE_KEY | |
271 | use_key = 2, | |
272 | use_key_begin = use_key | use_begin, | |
273 | use_key_end = use_key | use_end, | |
274 | use_key_first = use_key_begin | use_first, | |
275 | #endif | |
276 | use_data = 3, | |
277 | use_data_begin = use_data | use_begin, | |
278 | use_data_end = use_data | use_end, | |
279 | use_data_first = use_data_begin | use_first | |
280 | }; | |
281 | ||
282 | ||
283 | static int | |
284 | check_use (const char *data, nscd_ssize_t first_free, uint8_t *usemap, | |
285 | enum usekey use, ref_t start, size_t len) | |
286 | { | |
287 | assert (len >= 2); | |
288 | ||
289 | if (start > first_free || start + len > first_free | |
290 | || (start & BLOCK_ALIGN_M1)) | |
291 | return 0; | |
292 | ||
293 | if (usemap[start] == use_not) | |
294 | { | |
295 | /* Add the start marker. */ | |
296 | usemap[start] = use | use_begin; | |
297 | use &= ~use_first; | |
298 | ||
299 | while (--len > 0) | |
300 | if (usemap[++start] != use_not) | |
301 | return 0; | |
302 | else | |
303 | usemap[start] = use; | |
304 | ||
305 | /* Add the end marker. */ | |
306 | usemap[start] = use | use_end; | |
307 | } | |
308 | else if ((usemap[start] & ~use_first) == ((use | use_begin) & ~use_first)) | |
309 | { | |
310 | /* Hash entries can't be shared. */ | |
311 | if (use == use_he) | |
312 | return 0; | |
313 | ||
314 | usemap[start] |= (use & use_first); | |
315 | use &= ~use_first; | |
316 | ||
317 | while (--len > 1) | |
318 | if (usemap[++start] != use) | |
319 | return 0; | |
320 | ||
321 | if (usemap[++start] != (use | use_end)) | |
322 | return 0; | |
323 | } | |
324 | else | |
325 | /* Points to a wrong object or somewhere in the middle. */ | |
326 | return 0; | |
327 | ||
328 | return 1; | |
329 | } | |
330 | ||
331 | ||
332 | /* Verify data in persistent database. */ | |
333 | static int | |
334 | verify_persistent_db (void *mem, struct database_pers_head *readhead, int dbnr) | |
335 | { | |
b21fa963 | 336 | assert (dbnr == pwddb || dbnr == grpdb || dbnr == hstdb || dbnr == servdb); |
dc4bb1c2 UD |
337 | |
338 | time_t now = time (NULL); | |
339 | ||
340 | struct database_pers_head *head = mem; | |
341 | struct database_pers_head head_copy = *head; | |
342 | ||
343 | /* Check that the header that was read matches the head in the database. */ | |
344 | if (readhead != NULL && memcmp (head, readhead, sizeof (*head)) != 0) | |
345 | return 0; | |
346 | ||
347 | /* First some easy tests: make sure the database header is sane. */ | |
348 | if (head->version != DB_VERSION | |
349 | || head->header_size != sizeof (*head) | |
350 | /* We allow a timestamp to be one hour ahead of the current time. | |
351 | This should cover daylight saving time changes. */ | |
352 | || head->timestamp > now + 60 * 60 + 60 | |
353 | || (head->gc_cycle & 1) | |
354 | || (size_t) head->module > INT32_MAX / sizeof (ref_t) | |
355 | || (size_t) head->data_size > INT32_MAX - head->module * sizeof (ref_t) | |
356 | || head->first_free < 0 | |
357 | || head->first_free > head->data_size | |
358 | || (head->first_free & BLOCK_ALIGN_M1) != 0 | |
359 | || head->maxnentries < 0 | |
360 | || head->maxnsearched < 0) | |
361 | return 0; | |
362 | ||
363 | uint8_t *usemap = calloc (head->first_free, 1); | |
364 | if (usemap == NULL) | |
365 | return 0; | |
366 | ||
367 | const char *data = (char *) &head->array[roundup (head->module, | |
368 | ALIGN / sizeof (ref_t))]; | |
369 | ||
370 | nscd_ssize_t he_cnt = 0; | |
371 | for (nscd_ssize_t cnt = 0; cnt < head->module; ++cnt) | |
372 | { | |
373 | ref_t work = head->array[cnt]; | |
374 | ||
375 | while (work != ENDREF) | |
376 | { | |
377 | if (! check_use (data, head->first_free, usemap, use_he, work, | |
378 | sizeof (struct hashentry))) | |
379 | goto fail; | |
380 | ||
381 | /* Now we know we can dereference the record. */ | |
382 | struct hashentry *here = (struct hashentry *) (data + work); | |
383 | ||
384 | ++he_cnt; | |
385 | ||
386 | /* Make sure the record is for this type of service. */ | |
387 | if (here->type >= LASTREQ | |
388 | || serv2db[here->type] != &dbs[dbnr]) | |
389 | goto fail; | |
390 | ||
391 | /* Validate boolean field value. */ | |
392 | if (here->first != false && here->first != true) | |
393 | goto fail; | |
394 | ||
395 | if (here->len < 0) | |
396 | goto fail; | |
397 | ||
398 | /* Now the data. */ | |
399 | if (here->packet < 0 | |
400 | || here->packet > head->first_free | |
401 | || here->packet + sizeof (struct datahead) > head->first_free) | |
402 | goto fail; | |
403 | ||
404 | struct datahead *dh = (struct datahead *) (data + here->packet); | |
405 | ||
406 | if (! check_use (data, head->first_free, usemap, | |
407 | use_data | (here->first ? use_first : 0), | |
408 | here->packet, dh->allocsize)) | |
409 | goto fail; | |
410 | ||
411 | if (dh->allocsize < sizeof (struct datahead) | |
412 | || dh->recsize > dh->allocsize | |
413 | || (dh->notfound != false && dh->notfound != true) | |
414 | || (dh->usable != false && dh->usable != true)) | |
415 | goto fail; | |
416 | ||
417 | if (here->key < here->packet + sizeof (struct datahead) | |
418 | || here->key > here->packet + dh->allocsize | |
419 | || here->key + here->len > here->packet + dh->allocsize) | |
420 | { | |
421 | #if SEPARATE_KEY | |
422 | /* If keys can appear outside of data, this should be done | |
423 | instead. But gc doesn't mark the data in that case. */ | |
424 | if (! check_use (data, head->first_free, usemap, | |
425 | use_key | (here->first ? use_first : 0), | |
426 | here->key, here->len)) | |
427 | #endif | |
428 | goto fail; | |
429 | } | |
430 | ||
431 | work = here->next; | |
432 | } | |
433 | } | |
434 | ||
435 | if (he_cnt != head->nentries) | |
436 | goto fail; | |
437 | ||
438 | /* See if all data and keys had at least one reference from | |
439 | he->first == true hashentry. */ | |
440 | for (ref_t idx = 0; idx < head->first_free; ++idx) | |
441 | { | |
442 | #if SEPARATE_KEY | |
443 | if (usemap[idx] == use_key_begin) | |
444 | goto fail; | |
445 | #endif | |
446 | if (usemap[idx] == use_data_begin) | |
447 | goto fail; | |
448 | } | |
449 | ||
450 | /* Finally, make sure the database hasn't changed since the first test. */ | |
451 | if (memcmp (mem, &head_copy, sizeof (*head)) != 0) | |
452 | goto fail; | |
453 | ||
454 | free (usemap); | |
455 | return 1; | |
456 | ||
457 | fail: | |
458 | free (usemap); | |
459 | return 0; | |
460 | } | |
461 | ||
462 | ||
a334319f UD |
463 | /* Initialize database information structures. */ |
464 | void | |
465 | nscd_init (void) | |
0ecb606c | 466 | { |
057685e4 UD |
467 | /* Look up unprivileged uid/gid/groups before we start listening on the |
468 | socket */ | |
469 | if (server_user != NULL) | |
470 | begin_drop_privileges (); | |
471 | ||
67479a70 UD |
472 | if (nthreads == -1) |
473 | /* No configuration for this value, assume a default. */ | |
474 | nthreads = 2 * lastdb; | |
d67281a7 | 475 | |
d2dc7d84 | 476 | for (size_t cnt = 0; cnt < lastdb; ++cnt) |
67479a70 | 477 | if (dbs[cnt].enabled) |
9db29cde | 478 | { |
67479a70 | 479 | pthread_rwlock_init (&dbs[cnt].lock, NULL); |
a95a08b4 | 480 | pthread_mutex_init (&dbs[cnt].memlock, NULL); |
264d5b94 | 481 | |
a95a08b4 | 482 | if (dbs[cnt].persistent) |
e09edf23 | 483 | { |
a95a08b4 UD |
484 | /* Try to open the appropriate file on disk. */ |
485 | int fd = open (dbs[cnt].db_filename, O_RDWR); | |
486 | if (fd != -1) | |
487 | { | |
488 | struct stat64 st; | |
489 | void *mem; | |
490 | size_t total; | |
491 | struct database_pers_head head; | |
492 | ssize_t n = TEMP_FAILURE_RETRY (read (fd, &head, | |
493 | sizeof (head))); | |
494 | if (n != sizeof (head) || fstat64 (fd, &st) != 0) | |
495 | { | |
496 | fail_db: | |
497 | dbg_log (_("invalid persistent database file \"%s\": %s"), | |
498 | dbs[cnt].db_filename, strerror (errno)); | |
dc4bb1c2 | 499 | unlink (dbs[cnt].db_filename); |
a95a08b4 UD |
500 | } |
501 | else if (head.module == 0 && head.data_size == 0) | |
502 | { | |
503 | /* The file has been created, but the head has not been | |
504 | initialized yet. Remove the old file. */ | |
505 | unlink (dbs[cnt].db_filename); | |
506 | } | |
507 | else if (head.header_size != (int) sizeof (head)) | |
508 | { | |
509 | dbg_log (_("invalid persistent database file \"%s\": %s"), | |
510 | dbs[cnt].db_filename, | |
511 | _("header size does not match")); | |
dc4bb1c2 | 512 | unlink (dbs[cnt].db_filename); |
a95a08b4 UD |
513 | } |
514 | else if ((total = (sizeof (head) | |
c207f23b | 515 | + roundup (head.module * sizeof (ref_t), |
a95a08b4 UD |
516 | ALIGN) |
517 | + head.data_size)) | |
dc4bb1c2 UD |
518 | > st.st_size |
519 | || total < sizeof (head)) | |
a95a08b4 UD |
520 | { |
521 | dbg_log (_("invalid persistent database file \"%s\": %s"), | |
522 | dbs[cnt].db_filename, | |
523 | _("file size does not match")); | |
dc4bb1c2 | 524 | unlink (dbs[cnt].db_filename); |
a95a08b4 | 525 | } |
2c210d1e UD |
526 | /* Note we map with the maximum size allowed for the |
527 | database. This is likely much larger than the | |
528 | actual file size. This is OK on most OSes since | |
529 | extensions of the underlying file will | |
530 | automatically translate more pages available for | |
531 | memory access. */ | |
532 | else if ((mem = mmap (NULL, dbs[cnt].max_db_size, | |
533 | PROT_READ | PROT_WRITE, | |
534 | MAP_SHARED, fd, 0)) | |
535 | == MAP_FAILED) | |
a95a08b4 | 536 | goto fail_db; |
dc4bb1c2 UD |
537 | else if (!verify_persistent_db (mem, &head, cnt)) |
538 | { | |
539 | munmap (mem, total); | |
540 | dbg_log (_("invalid persistent database file \"%s\": %s"), | |
541 | dbs[cnt].db_filename, | |
542 | _("verification failed")); | |
543 | unlink (dbs[cnt].db_filename); | |
544 | } | |
a95a08b4 UD |
545 | else |
546 | { | |
547 | /* Success. We have the database. */ | |
548 | dbs[cnt].head = mem; | |
549 | dbs[cnt].memsize = total; | |
550 | dbs[cnt].data = (char *) | |
551 | &dbs[cnt].head->array[roundup (dbs[cnt].head->module, | |
552 | ALIGN / sizeof (ref_t))]; | |
553 | dbs[cnt].mmap_used = true; | |
554 | ||
555 | if (dbs[cnt].suggested_module > head.module) | |
556 | dbg_log (_("suggested size of table for database %s larger than the persistent database's table"), | |
557 | dbnames[cnt]); | |
558 | ||
559 | dbs[cnt].wr_fd = fd; | |
560 | fd = -1; | |
561 | /* We also need a read-only descriptor. */ | |
d13a3c57 UD |
562 | if (dbs[cnt].shared) |
563 | { | |
564 | dbs[cnt].ro_fd = open (dbs[cnt].db_filename, O_RDONLY); | |
565 | if (dbs[cnt].ro_fd == -1) | |
566 | dbg_log (_("\ | |
a95a08b4 | 567 | cannot create read-only descriptor for \"%s\"; no mmap"), |
d13a3c57 UD |
568 | dbs[cnt].db_filename); |
569 | } | |
a95a08b4 UD |
570 | |
571 | // XXX Shall we test whether the descriptors actually | |
572 | // XXX point to the same file? | |
573 | } | |
574 | ||
575 | /* Close the file descriptors in case something went | |
576 | wrong in which case the variable have not been | |
577 | assigned -1. */ | |
578 | if (fd != -1) | |
579 | close (fd); | |
580 | } | |
581 | } | |
582 | ||
583 | if (dbs[cnt].head == NULL) | |
584 | { | |
585 | /* No database loaded. Allocate the data structure, | |
586 | possibly on disk. */ | |
587 | struct database_pers_head head; | |
588 | size_t total = (sizeof (head) | |
589 | + roundup (dbs[cnt].suggested_module | |
590 | * sizeof (ref_t), ALIGN) | |
591 | + (dbs[cnt].suggested_module | |
592 | * DEFAULT_DATASIZE_PER_BUCKET)); | |
593 | ||
594 | /* Try to create the database. If we do not need a | |
595 | persistent database create a temporary file. */ | |
596 | int fd; | |
597 | int ro_fd = -1; | |
598 | if (dbs[cnt].persistent) | |
599 | { | |
600 | fd = open (dbs[cnt].db_filename, | |
601 | O_RDWR | O_CREAT | O_EXCL | O_TRUNC, | |
602 | S_IRUSR | S_IWUSR); | |
d13a3c57 | 603 | if (fd != -1 && dbs[cnt].shared) |
a95a08b4 UD |
604 | ro_fd = open (dbs[cnt].db_filename, O_RDONLY); |
605 | } | |
606 | else | |
607 | { | |
a8a58967 | 608 | char fname[] = _PATH_NSCD_XYZ_DB_TMP; |
a95a08b4 UD |
609 | fd = mkstemp (fname); |
610 | ||
611 | /* We do not need the file name anymore after we | |
612 | opened another file descriptor in read-only mode. */ | |
5ca3d19c | 613 | if (fd != -1) |
a95a08b4 | 614 | { |
5ca3d19c UD |
615 | if (dbs[cnt].shared) |
616 | ro_fd = open (fname, O_RDONLY); | |
a95a08b4 UD |
617 | |
618 | unlink (fname); | |
619 | } | |
620 | } | |
621 | ||
622 | if (fd == -1) | |
623 | { | |
624 | if (errno == EEXIST) | |
625 | { | |
626 | dbg_log (_("database for %s corrupted or simultaneously used; remove %s manually if necessary and restart"), | |
627 | dbnames[cnt], dbs[cnt].db_filename); | |
628 | // XXX Correct way to terminate? | |
629 | exit (1); | |
630 | } | |
631 | ||
632 | if (dbs[cnt].persistent) | |
633 | dbg_log (_("cannot create %s; no persistent database used"), | |
634 | dbs[cnt].db_filename); | |
635 | else | |
636 | dbg_log (_("cannot create %s; no sharing possible"), | |
637 | dbs[cnt].db_filename); | |
638 | ||
639 | dbs[cnt].persistent = 0; | |
640 | // XXX remember: no mmap | |
641 | } | |
642 | else | |
643 | { | |
644 | /* Tell the user if we could not create the read-only | |
645 | descriptor. */ | |
d13a3c57 | 646 | if (ro_fd == -1 && dbs[cnt].shared) |
a95a08b4 UD |
647 | dbg_log (_("\ |
648 | cannot create read-only descriptor for \"%s\"; no mmap"), | |
649 | dbs[cnt].db_filename); | |
650 | ||
651 | /* Before we create the header, initialiye the hash | |
652 | table. So that if we get interrupted if writing | |
653 | the header we can recognize a partially initialized | |
654 | database. */ | |
655 | size_t ps = sysconf (_SC_PAGESIZE); | |
656 | char tmpbuf[ps]; | |
657 | assert (~ENDREF == 0); | |
658 | memset (tmpbuf, '\xff', ps); | |
659 | ||
660 | size_t remaining = dbs[cnt].suggested_module * sizeof (ref_t); | |
661 | off_t offset = sizeof (head); | |
662 | ||
663 | size_t towrite; | |
664 | if (offset % ps != 0) | |
665 | { | |
666 | towrite = MIN (remaining, ps - (offset % ps)); | |
233399bc UD |
667 | if (pwrite (fd, tmpbuf, towrite, offset) != towrite) |
668 | goto write_fail; | |
a95a08b4 UD |
669 | offset += towrite; |
670 | remaining -= towrite; | |
671 | } | |
672 | ||
673 | while (remaining > ps) | |
674 | { | |
233399bc UD |
675 | if (pwrite (fd, tmpbuf, ps, offset) == -1) |
676 | goto write_fail; | |
a95a08b4 UD |
677 | offset += ps; |
678 | remaining -= ps; | |
679 | } | |
680 | ||
233399bc UD |
681 | if (remaining > 0 |
682 | && pwrite (fd, tmpbuf, remaining, offset) != remaining) | |
683 | goto write_fail; | |
a95a08b4 UD |
684 | |
685 | /* Create the header of the file. */ | |
686 | struct database_pers_head head = | |
687 | { | |
688 | .version = DB_VERSION, | |
689 | .header_size = sizeof (head), | |
690 | .module = dbs[cnt].suggested_module, | |
691 | .data_size = (dbs[cnt].suggested_module | |
692 | * DEFAULT_DATASIZE_PER_BUCKET), | |
693 | .first_free = 0 | |
694 | }; | |
695 | void *mem; | |
696 | ||
697 | if ((TEMP_FAILURE_RETRY (write (fd, &head, sizeof (head))) | |
698 | != sizeof (head)) | |
2c210d1e UD |
699 | || (TEMP_FAILURE_RETRY_VAL (posix_fallocate (fd, 0, total)) |
700 | != 0) | |
701 | || (mem = mmap (NULL, dbs[cnt].max_db_size, | |
702 | PROT_READ | PROT_WRITE, | |
a95a08b4 UD |
703 | MAP_SHARED, fd, 0)) == MAP_FAILED) |
704 | { | |
233399bc | 705 | write_fail: |
a95a08b4 UD |
706 | unlink (dbs[cnt].db_filename); |
707 | dbg_log (_("cannot write to database file %s: %s"), | |
708 | dbs[cnt].db_filename, strerror (errno)); | |
709 | dbs[cnt].persistent = 0; | |
710 | } | |
711 | else | |
712 | { | |
713 | /* Success. */ | |
714 | dbs[cnt].head = mem; | |
715 | dbs[cnt].data = (char *) | |
716 | &dbs[cnt].head->array[roundup (dbs[cnt].head->module, | |
717 | ALIGN / sizeof (ref_t))]; | |
718 | dbs[cnt].memsize = total; | |
719 | dbs[cnt].mmap_used = true; | |
720 | ||
721 | /* Remember the descriptors. */ | |
722 | dbs[cnt].wr_fd = fd; | |
723 | dbs[cnt].ro_fd = ro_fd; | |
724 | fd = -1; | |
725 | ro_fd = -1; | |
726 | } | |
727 | ||
728 | if (fd != -1) | |
729 | close (fd); | |
730 | if (ro_fd != -1) | |
731 | close (ro_fd); | |
732 | } | |
733 | } | |
734 | ||
4401d759 UD |
735 | if (paranoia |
736 | && ((dbs[cnt].wr_fd != -1 | |
737 | && fcntl (dbs[cnt].wr_fd, F_SETFD, FD_CLOEXEC) == -1) | |
738 | || (dbs[cnt].ro_fd != -1 | |
739 | && fcntl (dbs[cnt].ro_fd, F_SETFD, FD_CLOEXEC) == -1))) | |
740 | { | |
741 | dbg_log (_("\ | |
742 | cannot set socket to close on exec: %s; disabling paranoia mode"), | |
743 | strerror (errno)); | |
744 | paranoia = 0; | |
745 | } | |
746 | ||
a95a08b4 UD |
747 | if (dbs[cnt].head == NULL) |
748 | { | |
749 | /* We do not use the persistent database. Just | |
750 | create an in-memory data structure. */ | |
751 | assert (! dbs[cnt].persistent); | |
752 | ||
753 | dbs[cnt].head = xmalloc (sizeof (struct database_pers_head) | |
754 | + (dbs[cnt].suggested_module | |
755 | * sizeof (ref_t))); | |
25059769 | 756 | memset (dbs[cnt].head, '\0', sizeof (struct database_pers_head)); |
a95a08b4 UD |
757 | assert (~ENDREF == 0); |
758 | memset (dbs[cnt].head->array, '\xff', | |
759 | dbs[cnt].suggested_module * sizeof (ref_t)); | |
760 | dbs[cnt].head->module = dbs[cnt].suggested_module; | |
761 | dbs[cnt].head->data_size = (DEFAULT_DATASIZE_PER_BUCKET | |
762 | * dbs[cnt].head->module); | |
763 | dbs[cnt].data = xmalloc (dbs[cnt].head->data_size); | |
764 | dbs[cnt].head->first_free = 0; | |
c207f23b UD |
765 | |
766 | dbs[cnt].shared = 0; | |
767 | assert (dbs[cnt].ro_fd == -1); | |
e09edf23 | 768 | } |
d67281a7 | 769 | |
67479a70 | 770 | if (dbs[cnt].check_file) |
d67281a7 | 771 | { |
67479a70 | 772 | /* We need the modification date of the file. */ |
80ea3037 | 773 | struct stat64 st; |
d67281a7 | 774 | |
80ea3037 | 775 | if (stat64 (dbs[cnt].filename, &st) < 0) |
d67281a7 | 776 | { |
67479a70 UD |
777 | /* We cannot stat() the file, disable file checking. */ |
778 | dbg_log (_("cannot stat() file `%s': %s"), | |
e09edf23 | 779 | dbs[cnt].filename, strerror (errno)); |
67479a70 | 780 | dbs[cnt].check_file = 0; |
d67281a7 UD |
781 | } |
782 | else | |
67479a70 UD |
783 | dbs[cnt].file_mtime = st.st_mtime; |
784 | } | |
785 | } | |
d67281a7 UD |
786 | |
787 | /* Create the socket. */ | |
67479a70 UD |
788 | sock = socket (AF_UNIX, SOCK_STREAM, 0); |
789 | if (sock < 0) | |
d67281a7 | 790 | { |
67479a70 | 791 | dbg_log (_("cannot open socket: %s"), strerror (errno)); |
64d64de6 | 792 | exit (errno == EACCES ? 4 : 1); |
d67281a7 UD |
793 | } |
794 | /* Bind a name to the socket. */ | |
d2dc7d84 | 795 | struct sockaddr_un sock_addr; |
d67281a7 UD |
796 | sock_addr.sun_family = AF_UNIX; |
797 | strcpy (sock_addr.sun_path, _PATH_NSCDSOCKET); | |
67479a70 | 798 | if (bind (sock, (struct sockaddr *) &sock_addr, sizeof (sock_addr)) < 0) |
d67281a7 UD |
799 | { |
800 | dbg_log ("%s: %s", _PATH_NSCDSOCKET, strerror (errno)); | |
64d64de6 | 801 | exit (errno == EACCES ? 4 : 1); |
d67281a7 | 802 | } |
67479a70 | 803 | |
4401d759 | 804 | /* We don't want to get stuck on accept. */ |
d6db0975 | 805 | int fl = fcntl (sock, F_GETFL); |
4401d759 UD |
806 | if (fl == -1 || fcntl (sock, F_SETFL, fl | O_NONBLOCK) == -1) |
807 | { | |
808 | dbg_log (_("cannot change socket to nonblocking mode: %s"), | |
809 | strerror (errno)); | |
810 | exit (1); | |
811 | } | |
812 | ||
813 | /* The descriptor needs to be closed on exec. */ | |
814 | if (paranoia && fcntl (sock, F_SETFD, FD_CLOEXEC) == -1) | |
815 | { | |
816 | dbg_log (_("cannot set socket to close on exec: %s"), | |
817 | strerror (errno)); | |
818 | exit (1); | |
819 | } | |
d6db0975 | 820 | |
d67281a7 | 821 | /* Set permissions for the socket. */ |
a95a08b4 | 822 | chmod (_PATH_NSCDSOCKET, DEFFILEMODE); |
d67281a7 UD |
823 | |
824 | /* Set the socket up to accept connections. */ | |
67479a70 | 825 | if (listen (sock, SOMAXCONN) < 0) |
d67281a7 | 826 | { |
67479a70 UD |
827 | dbg_log (_("cannot enable socket to accept connections: %s"), |
828 | strerror (errno)); | |
d67281a7 UD |
829 | exit (1); |
830 | } | |
057685e4 UD |
831 | |
832 | /* Change to unprivileged uid/gid/groups if specifed in config file */ | |
833 | if (server_user != NULL) | |
834 | finish_drop_privileges (); | |
d67281a7 UD |
835 | } |
836 | ||
67479a70 UD |
837 | |
838 | /* Close the connections. */ | |
d67281a7 | 839 | void |
67479a70 | 840 | close_sockets (void) |
d67281a7 | 841 | { |
67479a70 UD |
842 | close (sock); |
843 | } | |
d67281a7 | 844 | |
a12ce44f | 845 | |
756409c4 | 846 | static void |
902c4291 | 847 | invalidate_cache (char *key, int fd) |
756409c4 UD |
848 | { |
849 | dbtype number; | |
902c4291 | 850 | int32_t resp; |
756409c4 | 851 | |
b21fa963 UD |
852 | for (number = pwddb; number < lastdb; ++number) |
853 | if (strcmp (key, dbnames[number]) == 0) | |
854 | { | |
855 | if (dbs[number].reset_res) | |
856 | res_init (); | |
482bbeb9 | 857 | |
b21fa963 UD |
858 | break; |
859 | } | |
860 | ||
861 | if (number == lastdb) | |
902c4291 UD |
862 | { |
863 | resp = EINVAL; | |
864 | writeall (fd, &resp, sizeof (resp)); | |
865 | return; | |
866 | } | |
756409c4 | 867 | |
fd665070 | 868 | if (dbs[number].enabled) |
902c4291 UD |
869 | prune_cache (&dbs[number], LONG_MAX, fd); |
870 | else | |
871 | { | |
872 | resp = 0; | |
873 | writeall (fd, &resp, sizeof (resp)); | |
874 | } | |
756409c4 UD |
875 | } |
876 | ||
67479a70 | 877 | |
c207f23b UD |
878 | #ifdef SCM_RIGHTS |
879 | static void | |
880 | send_ro_fd (struct database_dyn *db, char *key, int fd) | |
881 | { | |
882 | /* If we do not have an read-only file descriptor do nothing. */ | |
883 | if (db->ro_fd == -1) | |
884 | return; | |
885 | ||
886 | /* We need to send some data along with the descriptor. */ | |
887 | struct iovec iov[1]; | |
888 | iov[0].iov_base = key; | |
889 | iov[0].iov_len = strlen (key) + 1; | |
890 | ||
891 | /* Prepare the control message to transfer the descriptor. */ | |
a08ab897 UD |
892 | union |
893 | { | |
894 | struct cmsghdr hdr; | |
895 | char bytes[CMSG_SPACE (sizeof (int))]; | |
896 | } buf; | |
c207f23b | 897 | struct msghdr msg = { .msg_iov = iov, .msg_iovlen = 1, |
a08ab897 UD |
898 | .msg_control = buf.bytes, |
899 | .msg_controllen = sizeof (buf) }; | |
c207f23b UD |
900 | struct cmsghdr *cmsg = CMSG_FIRSTHDR (&msg); |
901 | ||
902 | cmsg->cmsg_level = SOL_SOCKET; | |
903 | cmsg->cmsg_type = SCM_RIGHTS; | |
904 | cmsg->cmsg_len = CMSG_LEN (sizeof (int)); | |
905 | ||
906 | *(int *) CMSG_DATA (cmsg) = db->ro_fd; | |
907 | ||
908 | msg.msg_controllen = cmsg->cmsg_len; | |
909 | ||
910 | /* Send the control message. We repeat when we are interrupted but | |
911 | everything else is ignored. */ | |
6925ef9a UD |
912 | #ifndef MSG_NOSIGNAL |
913 | # define MSG_NOSIGNAL 0 | |
914 | #endif | |
915 | (void) TEMP_FAILURE_RETRY (sendmsg (fd, &msg, MSG_NOSIGNAL)); | |
c207f23b UD |
916 | |
917 | if (__builtin_expect (debug_level > 0, 0)) | |
918 | dbg_log (_("provide access to FD %d, for %s"), db->ro_fd, key); | |
919 | } | |
920 | #endif /* SCM_RIGHTS */ | |
921 | ||
922 | ||
67479a70 UD |
923 | /* Handle new request. */ |
924 | static void | |
a1c542bf | 925 | handle_request (int fd, request_header *req, void *key, uid_t uid) |
67479a70 | 926 | { |
23700036 | 927 | if (__builtin_expect (req->version, NSCD_VERSION) != NSCD_VERSION) |
d67281a7 | 928 | { |
98e75a1c UD |
929 | if (debug_level > 0) |
930 | dbg_log (_("\ | |
67479a70 | 931 | cannot handle old request version %d; current version is %d"), |
98e75a1c | 932 | req->version, NSCD_VERSION); |
d67281a7 UD |
933 | return; |
934 | } | |
935 | ||
74a30a58 UD |
936 | /* Make the SELinux check before we go on to the standard checks. We |
937 | need to verify that the request type is valid, since it has not | |
938 | yet been checked at this point. */ | |
939 | if (selinux_enabled | |
43397eaf UD |
940 | && __builtin_expect (req->type >= GETPWBYNAME, 1) |
941 | && __builtin_expect (req->type < LASTREQ, 1) | |
74a30a58 UD |
942 | && nscd_request_avc_has_perm (fd, req->type) != 0) |
943 | return; | |
944 | ||
a95a08b4 UD |
945 | struct database_dyn *db = serv2db[req->type]; |
946 | ||
f7e7a396 UD |
947 | // XXX Clean up so that each new command need not introduce a |
948 | // XXX new conditional. | |
d19687d6 | 949 | if ((__builtin_expect (req->type, GETPWBYNAME) >= GETPWBYNAME |
b21fa963 UD |
950 | && __builtin_expect (req->type, GETHOSTBYADDRv6) <= GETHOSTBYADDRv6) |
951 | || req->type == GETAI || req->type == INITGROUPS | |
952 | || req->type == GETSERVBYNAME || req->type == GETSERVBYPORT) | |
d67281a7 | 953 | { |
23700036 | 954 | if (__builtin_expect (debug_level, 0) > 0) |
8d8c6efa UD |
955 | { |
956 | if (req->type == GETHOSTBYADDR || req->type == GETHOSTBYADDRv6) | |
957 | { | |
958 | char buf[INET6_ADDRSTRLEN]; | |
959 | ||
960 | dbg_log ("\t%s (%s)", serv2str[req->type], | |
961 | inet_ntop (req->type == GETHOSTBYADDR | |
962 | ? AF_INET : AF_INET6, | |
963 | key, buf, sizeof (buf))); | |
964 | } | |
965 | else | |
a95a08b4 | 966 | dbg_log ("\t%s (%s)", serv2str[req->type], (char *) key); |
8d8c6efa | 967 | } |
d67281a7 | 968 | |
67479a70 UD |
969 | /* Is this service enabled? */ |
970 | if (!db->enabled) | |
971 | { | |
ce85d65b | 972 | /* No, sent the prepared record. */ |
2c210d1e UD |
973 | if (TEMP_FAILURE_RETRY (send (fd, db->disabled_iov->iov_base, |
974 | db->disabled_iov->iov_len, | |
975 | MSG_NOSIGNAL)) | |
4c5dd2a2 | 976 | != (ssize_t) db->disabled_iov->iov_len |
23700036 | 977 | && __builtin_expect (debug_level, 0) > 0) |
67479a70 UD |
978 | { |
979 | /* We have problems sending the result. */ | |
980 | char buf[256]; | |
981 | dbg_log (_("cannot write result: %s"), | |
982 | strerror_r (errno, buf, sizeof (buf))); | |
983 | } | |
d67281a7 | 984 | |
67479a70 UD |
985 | return; |
986 | } | |
d67281a7 | 987 | |
67479a70 | 988 | /* Be sure we can read the data. */ |
c86e6aec UD |
989 | if (__builtin_expect (pthread_rwlock_tryrdlock (&db->lock) != 0, 0)) |
990 | { | |
a95a08b4 | 991 | ++db->head->rdlockdelayed; |
c86e6aec UD |
992 | pthread_rwlock_rdlock (&db->lock); |
993 | } | |
67479a70 UD |
994 | |
995 | /* See whether we can handle it from the cache. */ | |
a95a08b4 UD |
996 | struct datahead *cached; |
997 | cached = (struct datahead *) cache_search (req->type, key, req->key_len, | |
998 | db, uid); | |
67479a70 UD |
999 | if (cached != NULL) |
1000 | { | |
1001 | /* Hurray it's in the cache. */ | |
eac10791 UD |
1002 | ssize_t nwritten; |
1003 | ||
1004 | #ifdef HAVE_SENDFILE | |
1005 | if (db->mmap_used || !cached->notfound) | |
1006 | { | |
1007 | assert (db->wr_fd != -1); | |
1008 | assert ((char *) cached->data > (char *) db->data); | |
1009 | assert ((char *) cached->data - (char *) db->head | |
1010 | + cached->recsize | |
1011 | <= (sizeof (struct database_pers_head) | |
1012 | + db->head->module * sizeof (ref_t) | |
1013 | + db->head->data_size)); | |
bd547139 UD |
1014 | nwritten = sendfileall (fd, db->wr_fd, |
1015 | (char *) cached->data | |
1016 | - (char *) db->head, cached->recsize); | |
eac10791 UD |
1017 | # ifndef __ASSUME_SENDFILE |
1018 | if (nwritten == -1 && errno == ENOSYS) | |
1019 | goto use_write; | |
1020 | # endif | |
1021 | } | |
1022 | else | |
1023 | # ifndef __ASSUME_SENDFILE | |
1024 | use_write: | |
1025 | # endif | |
1026 | #endif | |
1027 | nwritten = writeall (fd, cached->data, cached->recsize); | |
1028 | ||
1029 | if (nwritten != cached->recsize | |
23700036 | 1030 | && __builtin_expect (debug_level, 0) > 0) |
67479a70 UD |
1031 | { |
1032 | /* We have problems sending the result. */ | |
1033 | char buf[256]; | |
1034 | dbg_log (_("cannot write result: %s"), | |
1035 | strerror_r (errno, buf, sizeof (buf))); | |
1036 | } | |
1037 | ||
1038 | pthread_rwlock_unlock (&db->lock); | |
1039 | ||
1040 | return; | |
1041 | } | |
1042 | ||
1043 | pthread_rwlock_unlock (&db->lock); | |
d67281a7 | 1044 | } |
23700036 | 1045 | else if (__builtin_expect (debug_level, 0) > 0) |
756409c4 UD |
1046 | { |
1047 | if (req->type == INVALIDATE) | |
c207f23b | 1048 | dbg_log ("\t%s (%s)", serv2str[req->type], (char *) key); |
0532e21b | 1049 | else |
a95a08b4 | 1050 | dbg_log ("\t%s", serv2str[req->type]); |
756409c4 | 1051 | } |
67479a70 UD |
1052 | |
1053 | /* Handle the request. */ | |
1054 | switch (req->type) | |
d67281a7 | 1055 | { |
67479a70 | 1056 | case GETPWBYNAME: |
a95a08b4 | 1057 | addpwbyname (db, fd, req, key, uid); |
67479a70 UD |
1058 | break; |
1059 | ||
1060 | case GETPWBYUID: | |
a95a08b4 | 1061 | addpwbyuid (db, fd, req, key, uid); |
67479a70 UD |
1062 | break; |
1063 | ||
1064 | case GETGRBYNAME: | |
a95a08b4 | 1065 | addgrbyname (db, fd, req, key, uid); |
67479a70 UD |
1066 | break; |
1067 | ||
1068 | case GETGRBYGID: | |
a95a08b4 | 1069 | addgrbygid (db, fd, req, key, uid); |
67479a70 UD |
1070 | break; |
1071 | ||
1072 | case GETHOSTBYNAME: | |
a95a08b4 | 1073 | addhstbyname (db, fd, req, key, uid); |
67479a70 UD |
1074 | break; |
1075 | ||
1076 | case GETHOSTBYNAMEv6: | |
a95a08b4 | 1077 | addhstbynamev6 (db, fd, req, key, uid); |
67479a70 UD |
1078 | break; |
1079 | ||
1080 | case GETHOSTBYADDR: | |
a95a08b4 | 1081 | addhstbyaddr (db, fd, req, key, uid); |
67479a70 UD |
1082 | break; |
1083 | ||
1084 | case GETHOSTBYADDRv6: | |
a95a08b4 | 1085 | addhstbyaddrv6 (db, fd, req, key, uid); |
67479a70 UD |
1086 | break; |
1087 | ||
d19687d6 UD |
1088 | case GETAI: |
1089 | addhstai (db, fd, req, key, uid); | |
1090 | break; | |
1091 | ||
f7e7a396 UD |
1092 | case INITGROUPS: |
1093 | addinitgroups (db, fd, req, key, uid); | |
1094 | break; | |
1095 | ||
b21fa963 UD |
1096 | case GETSERVBYNAME: |
1097 | addservbyname (db, fd, req, key, uid); | |
1098 | break; | |
1099 | ||
1100 | case GETSERVBYPORT: | |
1101 | addservbyport (db, fd, req, key, uid); | |
1102 | break; | |
1103 | ||
67479a70 | 1104 | case GETSTAT: |
67479a70 | 1105 | case SHUTDOWN: |
756409c4 | 1106 | case INVALIDATE: |
70e2ebba UD |
1107 | { |
1108 | /* Get the callers credentials. */ | |
cedc8559 | 1109 | #ifdef SO_PEERCRED |
70e2ebba UD |
1110 | struct ucred caller; |
1111 | socklen_t optlen = sizeof (caller); | |
be3c40b6 | 1112 | |
70e2ebba UD |
1113 | if (getsockopt (fd, SOL_SOCKET, SO_PEERCRED, &caller, &optlen) < 0) |
1114 | { | |
1115 | char buf[256]; | |
a1c542bf | 1116 | |
c69136ae | 1117 | dbg_log (_("error getting caller's id: %s"), |
70e2ebba UD |
1118 | strerror_r (errno, buf, sizeof (buf))); |
1119 | break; | |
1120 | } | |
a12ce44f | 1121 | |
70e2ebba | 1122 | uid = caller.uid; |
a12ce44f | 1123 | #else |
70e2ebba UD |
1124 | /* Some systems have no SO_PEERCRED implementation. They don't |
1125 | care about security so we don't as well. */ | |
1126 | uid = 0; | |
cedc8559 | 1127 | #endif |
70e2ebba | 1128 | } |
a12ce44f UD |
1129 | |
1130 | /* Accept shutdown, getstat and invalidate only from root. For | |
1131 | the stat call also allow the user specified in the config file. */ | |
1132 | if (req->type == GETSTAT) | |
1133 | { | |
1134 | if (uid == 0 || uid == stat_uid) | |
1135 | send_stats (fd, dbs); | |
1136 | } | |
1137 | else if (uid == 0) | |
1138 | { | |
1139 | if (req->type == INVALIDATE) | |
902c4291 | 1140 | invalidate_cache (key, fd); |
a12ce44f UD |
1141 | else |
1142 | termination_handler (0); | |
a1c542bf | 1143 | } |
67479a70 UD |
1144 | break; |
1145 | ||
c207f23b UD |
1146 | case GETFDPW: |
1147 | case GETFDGR: | |
1148 | case GETFDHST: | |
b21fa963 | 1149 | case GETFDSERV: |
c207f23b UD |
1150 | #ifdef SCM_RIGHTS |
1151 | send_ro_fd (serv2db[req->type], key, fd); | |
1152 | #endif | |
1153 | break; | |
1154 | ||
67479a70 | 1155 | default: |
64acf8ed UD |
1156 | /* Ignore the command, it's nothing we know. */ |
1157 | break; | |
d67281a7 | 1158 | } |
67479a70 UD |
1159 | } |
1160 | ||
d67281a7 | 1161 | |
4401d759 UD |
1162 | /* Restart the process. */ |
1163 | static void | |
1164 | restart (void) | |
1165 | { | |
1166 | /* First determine the parameters. We do not use the parameters | |
1167 | passed to main() since in case nscd is started by running the | |
1168 | dynamic linker this will not work. Yes, this is not the usual | |
1169 | case but nscd is part of glibc and we occasionally do this. */ | |
1170 | size_t buflen = 1024; | |
1171 | char *buf = alloca (buflen); | |
1172 | size_t readlen = 0; | |
1173 | int fd = open ("/proc/self/cmdline", O_RDONLY); | |
1174 | if (fd == -1) | |
1175 | { | |
1176 | dbg_log (_("\ | |
1177 | cannot open /proc/self/cmdline: %s; disabling paranoia mode"), | |
1178 | strerror (errno)); | |
1179 | ||
1180 | paranoia = 0; | |
1181 | return; | |
1182 | } | |
1183 | ||
1184 | while (1) | |
1185 | { | |
1186 | ssize_t n = TEMP_FAILURE_RETRY (read (fd, buf + readlen, | |
1187 | buflen - readlen)); | |
1188 | if (n == -1) | |
1189 | { | |
1190 | dbg_log (_("\ | |
b21fa963 | 1191 | cannot read /proc/self/cmdline: %s; disabling paranoia mode"), |
4401d759 UD |
1192 | strerror (errno)); |
1193 | ||
1194 | close (fd); | |
1195 | paranoia = 0; | |
1196 | return; | |
1197 | } | |
1198 | ||
1199 | readlen += n; | |
1200 | ||
1201 | if (readlen < buflen) | |
1202 | break; | |
1203 | ||
1204 | /* We might have to extend the buffer. */ | |
1205 | size_t old_buflen = buflen; | |
1206 | char *newp = extend_alloca (buf, buflen, 2 * buflen); | |
1207 | buf = memmove (newp, buf, old_buflen); | |
1208 | } | |
1209 | ||
1210 | close (fd); | |
1211 | ||
1212 | /* Parse the command line. Worst case scenario: every two | |
1213 | characters form one parameter (one character plus NUL). */ | |
1214 | char **argv = alloca ((readlen / 2 + 1) * sizeof (argv[0])); | |
1215 | int argc = 0; | |
1216 | ||
1217 | char *cp = buf; | |
1218 | while (cp < buf + readlen) | |
1219 | { | |
1220 | argv[argc++] = cp; | |
1221 | cp = (char *) rawmemchr (cp, '\0') + 1; | |
1222 | } | |
1223 | argv[argc] = NULL; | |
1224 | ||
1225 | /* Second, change back to the old user if we changed it. */ | |
1226 | if (server_user != NULL) | |
1227 | { | |
464c9fad | 1228 | if (setresuid (old_uid, old_uid, old_uid) != 0) |
4401d759 UD |
1229 | { |
1230 | dbg_log (_("\ | |
1231 | cannot change to old UID: %s; disabling paranoia mode"), | |
1232 | strerror (errno)); | |
1233 | ||
1234 | paranoia = 0; | |
1235 | return; | |
1236 | } | |
1237 | ||
464c9fad | 1238 | if (setresgid (old_gid, old_gid, old_gid) != 0) |
4401d759 UD |
1239 | { |
1240 | dbg_log (_("\ | |
1241 | cannot change to old GID: %s; disabling paranoia mode"), | |
1242 | strerror (errno)); | |
1243 | ||
1244 | setuid (server_uid); | |
1245 | paranoia = 0; | |
1246 | return; | |
1247 | } | |
1248 | } | |
1249 | ||
1250 | /* Next change back to the old working directory. */ | |
1251 | if (chdir (oldcwd) == -1) | |
1252 | { | |
1253 | dbg_log (_("\ | |
1254 | cannot change to old working directory: %s; disabling paranoia mode"), | |
1255 | strerror (errno)); | |
1256 | ||
1257 | if (server_user != NULL) | |
1258 | { | |
1259 | setuid (server_uid); | |
1260 | setgid (server_gid); | |
1261 | } | |
1262 | paranoia = 0; | |
1263 | return; | |
1264 | } | |
1265 | ||
1266 | /* Synchronize memory. */ | |
1267 | for (int cnt = 0; cnt < lastdb; ++cnt) | |
1268 | { | |
1269 | /* Make sure nobody keeps using the database. */ | |
1270 | dbs[cnt].head->timestamp = 0; | |
1271 | ||
1272 | if (dbs[cnt].persistent) | |
1273 | // XXX async OK? | |
1274 | msync (dbs[cnt].head, dbs[cnt].memsize, MS_ASYNC); | |
1275 | } | |
1276 | ||
1277 | /* The preparations are done. */ | |
1278 | execv ("/proc/self/exe", argv); | |
1279 | ||
1280 | /* If we come here, we will never be able to re-exec. */ | |
1281 | dbg_log (_("re-exec failed: %s; disabling paranoia mode"), | |
1282 | strerror (errno)); | |
1283 | ||
1284 | if (server_user != NULL) | |
1285 | { | |
1286 | setuid (server_uid); | |
1287 | setgid (server_gid); | |
1288 | } | |
233399bc UD |
1289 | if (chdir ("/") != 0) |
1290 | dbg_log (_("cannot change current working directory to \"/\": %s"), | |
1291 | strerror (errno)); | |
4401d759 UD |
1292 | paranoia = 0; |
1293 | } | |
1294 | ||
1295 | ||
1945c96f UD |
1296 | /* List of file descriptors. */ |
1297 | struct fdlist | |
1298 | { | |
1299 | int fd; | |
1300 | struct fdlist *next; | |
1301 | }; | |
1302 | /* Memory allocated for the list. */ | |
1303 | static struct fdlist *fdlist; | |
1304 | /* List of currently ready-to-read file descriptors. */ | |
1305 | static struct fdlist *readylist; | |
1306 | ||
1307 | /* Conditional variable and mutex to signal availability of entries in | |
1308 | READYLIST. The condvar is initialized dynamically since we might | |
1309 | use a different clock depending on availability. */ | |
1310 | static pthread_cond_t readylist_cond; | |
1311 | static pthread_mutex_t readylist_lock = PTHREAD_MUTEX_INITIALIZER; | |
1312 | ||
1313 | /* The clock to use with the condvar. */ | |
1314 | static clockid_t timeout_clock = CLOCK_REALTIME; | |
1315 | ||
1316 | /* Number of threads ready to handle the READYLIST. */ | |
1317 | static unsigned long int nready; | |
1318 | ||
1319 | ||
67479a70 UD |
1320 | /* This is the main loop. It is replicated in different threads but the |
1321 | `poll' call makes sure only one thread handles an incoming connection. */ | |
1322 | static void * | |
1323 | __attribute__ ((__noreturn__)) | |
1324 | nscd_run (void *p) | |
1325 | { | |
1945c96f UD |
1326 | const long int my_number = (long int) p; |
1327 | const int run_prune = my_number < lastdb && dbs[my_number].enabled; | |
1328 | struct timespec prune_ts; | |
1329 | int to = 0; | |
1330 | char buf[256]; | |
264d5b94 | 1331 | |
72ae1e97 | 1332 | if (run_prune) |
1945c96f UD |
1333 | { |
1334 | setup_thread (&dbs[my_number]); | |
081fc592 | 1335 | |
1945c96f UD |
1336 | /* We are running. */ |
1337 | dbs[my_number].head->timestamp = time (NULL); | |
d67281a7 | 1338 | |
1945c96f UD |
1339 | if (clock_gettime (timeout_clock, &prune_ts) == -1) |
1340 | /* Should never happen. */ | |
1341 | abort (); | |
0fdb4f42 | 1342 | |
1945c96f UD |
1343 | /* Compute timeout time. */ |
1344 | prune_ts.tv_sec += CACHE_PRUNE_INTERVAL; | |
1345 | } | |
1346 | ||
1347 | /* Initial locking. */ | |
1348 | pthread_mutex_lock (&readylist_lock); | |
1349 | ||
1350 | /* One more thread available. */ | |
1351 | ++nready; | |
0fdb4f42 | 1352 | |
1945c96f UD |
1353 | while (1) |
1354 | { | |
1355 | while (readylist == NULL) | |
d6db0975 | 1356 | { |
d6db0975 UD |
1357 | if (run_prune) |
1358 | { | |
1945c96f UD |
1359 | /* Wait, but not forever. */ |
1360 | to = pthread_cond_timedwait (&readylist_cond, &readylist_lock, | |
1361 | &prune_ts); | |
1362 | ||
1363 | /* If we were woken and there is no work to be done, | |
1364 | just start pruning. */ | |
1365 | if (readylist == NULL && to == ETIMEDOUT) | |
1366 | { | |
4401d759 | 1367 | --nready; |
1945c96f UD |
1368 | pthread_mutex_unlock (&readylist_lock); |
1369 | goto only_prune; | |
1370 | } | |
d6db0975 | 1371 | } |
1945c96f UD |
1372 | else |
1373 | /* No need to timeout. */ | |
1374 | pthread_cond_wait (&readylist_cond, &readylist_lock); | |
1375 | } | |
0fdb4f42 | 1376 | |
1945c96f UD |
1377 | struct fdlist *it = readylist->next; |
1378 | if (readylist->next == readylist) | |
1379 | /* Just one entry on the list. */ | |
1380 | readylist = NULL; | |
1381 | else | |
1382 | readylist->next = it->next; | |
0fdb4f42 | 1383 | |
1945c96f UD |
1384 | /* Extract the information and mark the record ready to be used |
1385 | again. */ | |
1386 | int fd = it->fd; | |
1387 | it->next = NULL; | |
0fdb4f42 | 1388 | |
1945c96f UD |
1389 | /* One more thread available. */ |
1390 | --nready; | |
67479a70 | 1391 | |
1945c96f UD |
1392 | /* We are done with the list. */ |
1393 | pthread_mutex_unlock (&readylist_lock); | |
67479a70 | 1394 | |
1945c96f UD |
1395 | /* We do not want to block on a short read or so. */ |
1396 | int fl = fcntl (fd, F_GETFL); | |
1397 | if (fl == -1 || fcntl (fd, F_SETFL, fl | O_NONBLOCK) == -1) | |
1398 | goto close_and_out; | |
0fdb4f42 UD |
1399 | |
1400 | /* Now read the request. */ | |
1945c96f | 1401 | request_header req; |
0fdb4f42 UD |
1402 | if (__builtin_expect (TEMP_FAILURE_RETRY (read (fd, &req, sizeof (req))) |
1403 | != sizeof (req), 0)) | |
d67281a7 | 1404 | { |
1945c96f UD |
1405 | /* We failed to read data. Note that this also might mean we |
1406 | failed because we would have blocked. */ | |
0fdb4f42 UD |
1407 | if (debug_level > 0) |
1408 | dbg_log (_("short read while reading request: %s"), | |
1409 | strerror_r (errno, buf, sizeof (buf))); | |
1945c96f | 1410 | goto close_and_out; |
0fdb4f42 UD |
1411 | } |
1412 | ||
3c82c131 UD |
1413 | /* Check whether this is a valid request type. */ |
1414 | if (req.type < GETPWBYNAME || req.type >= LASTREQ) | |
1415 | goto close_and_out; | |
1416 | ||
0fdb4f42 UD |
1417 | /* Some systems have no SO_PEERCRED implementation. They don't |
1418 | care about security so we don't as well. */ | |
1945c96f | 1419 | uid_t uid = -1; |
c86e6aec | 1420 | #ifdef SO_PEERCRED |
1945c96f UD |
1421 | pid_t pid = 0; |
1422 | ||
70e2ebba | 1423 | if (__builtin_expect (debug_level > 0, 0)) |
0fdb4f42 UD |
1424 | { |
1425 | struct ucred caller; | |
1426 | socklen_t optlen = sizeof (caller); | |
c86e6aec | 1427 | |
0fdb4f42 UD |
1428 | if (getsockopt (fd, SOL_SOCKET, SO_PEERCRED, &caller, &optlen) == 0) |
1429 | pid = caller.pid; | |
1430 | } | |
cedc8559 | 1431 | #endif |
a1c542bf | 1432 | |
0fdb4f42 UD |
1433 | /* It should not be possible to crash the nscd with a silly |
1434 | request (i.e., a terribly large key). We limit the size to 1kb. */ | |
1945c96f | 1435 | #define MAXKEYLEN 1024 |
0fdb4f42 | 1436 | if (__builtin_expect (req.key_len, 1) < 0 |
1945c96f | 1437 | || __builtin_expect (req.key_len, 1) > MAXKEYLEN) |
0fdb4f42 UD |
1438 | { |
1439 | if (debug_level > 0) | |
1440 | dbg_log (_("key length in request too long: %d"), req.key_len); | |
0fdb4f42 UD |
1441 | } |
1442 | else | |
1443 | { | |
1444 | /* Get the key. */ | |
1945c96f | 1445 | char keybuf[MAXKEYLEN]; |
0fdb4f42 UD |
1446 | |
1447 | if (__builtin_expect (TEMP_FAILURE_RETRY (read (fd, keybuf, | |
1448 | req.key_len)) | |
1449 | != req.key_len, 0)) | |
67479a70 | 1450 | { |
1945c96f | 1451 | /* Again, this can also mean we would have blocked. */ |
98e75a1c | 1452 | if (debug_level > 0) |
0fdb4f42 UD |
1453 | dbg_log (_("short read while reading request key: %s"), |
1454 | strerror_r (errno, buf, sizeof (buf))); | |
1945c96f | 1455 | goto close_and_out; |
67479a70 | 1456 | } |
0fdb4f42 UD |
1457 | |
1458 | if (__builtin_expect (debug_level, 0) > 0) | |
67479a70 | 1459 | { |
c86e6aec | 1460 | #ifdef SO_PEERCRED |
0fdb4f42 UD |
1461 | if (pid != 0) |
1462 | dbg_log (_("\ | |
c86e6aec | 1463 | handle_request: request received (Version = %d) from PID %ld"), |
0fdb4f42 UD |
1464 | req.version, (long int) pid); |
1465 | else | |
c86e6aec | 1466 | #endif |
0fdb4f42 | 1467 | dbg_log (_("\ |
c86e6aec | 1468 | handle_request: request received (Version = %d)"), req.version); |
0fdb4f42 | 1469 | } |
c86e6aec | 1470 | |
0fdb4f42 UD |
1471 | /* Phew, we got all the data, now process it. */ |
1472 | handle_request (fd, &req, keybuf, uid); | |
d67281a7 | 1473 | } |
264d5b94 | 1474 | |
3c82c131 UD |
1475 | close_and_out: |
1476 | /* We are done. */ | |
1477 | close (fd); | |
1478 | ||
1945c96f UD |
1479 | /* Check whether we should be pruning the cache. */ |
1480 | assert (run_prune || to == 0); | |
1481 | if (to == ETIMEDOUT) | |
1482 | { | |
1483 | only_prune: | |
1484 | /* The pthread_cond_timedwait() call timed out. It is time | |
1485 | to clean up the cache. */ | |
1486 | assert (my_number < lastdb); | |
902c4291 | 1487 | prune_cache (&dbs[my_number], time (NULL), -1); |
1945c96f UD |
1488 | |
1489 | if (clock_gettime (timeout_clock, &prune_ts) == -1) | |
1490 | /* Should never happen. */ | |
1491 | abort (); | |
1492 | ||
1493 | /* Compute next timeout time. */ | |
1494 | prune_ts.tv_sec += CACHE_PRUNE_INTERVAL; | |
1495 | ||
1496 | /* In case the list is emtpy we do not want to run the prune | |
1497 | code right away again. */ | |
1498 | to = 0; | |
1499 | } | |
1500 | ||
1501 | /* Re-locking. */ | |
1502 | pthread_mutex_lock (&readylist_lock); | |
1503 | ||
1504 | /* One more thread available. */ | |
1505 | ++nready; | |
1506 | } | |
1507 | } | |
1508 | ||
1509 | ||
fc03df7a UD |
1510 | static unsigned int nconns; |
1511 | ||
1945c96f | 1512 | static void |
fc03df7a | 1513 | fd_ready (int fd) |
1945c96f | 1514 | { |
fc03df7a UD |
1515 | pthread_mutex_lock (&readylist_lock); |
1516 | ||
1517 | /* Find an empty entry in FDLIST. */ | |
1518 | size_t inner; | |
1519 | for (inner = 0; inner < nconns; ++inner) | |
1520 | if (fdlist[inner].next == NULL) | |
1521 | break; | |
1522 | assert (inner < nconns); | |
1523 | ||
1524 | fdlist[inner].fd = fd; | |
1525 | ||
1526 | if (readylist == NULL) | |
1527 | readylist = fdlist[inner].next = &fdlist[inner]; | |
1945c96f | 1528 | else |
fc03df7a UD |
1529 | { |
1530 | fdlist[inner].next = readylist->next; | |
1531 | readylist = readylist->next = &fdlist[inner]; | |
1532 | } | |
1533 | ||
1534 | bool do_signal = true; | |
1535 | if (__builtin_expect (nready == 0, 0)) | |
1536 | { | |
1537 | ++client_queued; | |
1538 | do_signal = false; | |
27e82856 UD |
1539 | |
1540 | /* Try to start another thread to help out. */ | |
1541 | pthread_t th; | |
1542 | if (nthreads < max_nthreads | |
1543 | && pthread_create (&th, &attr, nscd_run, | |
1544 | (void *) (long int) nthreads) == 0) | |
1545 | { | |
1546 | /* We got another thread. */ | |
1547 | ++nthreads; | |
908c9e87 | 1548 | /* The new thread might need a kick. */ |
27e82856 UD |
1549 | do_signal = true; |
1550 | } | |
1551 | ||
fc03df7a UD |
1552 | } |
1553 | ||
1554 | pthread_mutex_unlock (&readylist_lock); | |
1555 | ||
1556 | /* Tell one of the worker threads there is work to do. */ | |
1557 | if (do_signal) | |
1558 | pthread_cond_signal (&readylist_cond); | |
1559 | } | |
1945c96f | 1560 | |
fc03df7a | 1561 | |
4401d759 UD |
1562 | /* Check whether restarting should happen. */ |
1563 | static inline int | |
1564 | restart_p (time_t now) | |
1565 | { | |
1566 | return (paranoia && readylist == NULL && nready == nthreads | |
1567 | && now >= restart_time); | |
1568 | } | |
1569 | ||
1570 | ||
1571 | /* Array for times a connection was accepted. */ | |
fc03df7a UD |
1572 | static time_t *starttime; |
1573 | ||
1574 | ||
1575 | static void | |
1576 | __attribute__ ((__noreturn__)) | |
1577 | main_loop_poll (void) | |
1578 | { | |
1945c96f UD |
1579 | struct pollfd *conns = (struct pollfd *) xmalloc (nconns |
1580 | * sizeof (conns[0])); | |
1581 | ||
1945c96f UD |
1582 | conns[0].fd = sock; |
1583 | conns[0].events = POLLRDNORM; | |
1584 | size_t nused = 1; | |
1585 | size_t firstfree = 1; | |
1586 | ||
1587 | while (1) | |
1588 | { | |
1589 | /* Wait for any event. We wait at most a couple of seconds so | |
1590 | that we can check whether we should close any of the accepted | |
1591 | connections since we have not received a request. */ | |
1592 | #define MAX_ACCEPT_TIMEOUT 30 | |
1593 | #define MIN_ACCEPT_TIMEOUT 5 | |
1594 | #define MAIN_THREAD_TIMEOUT \ | |
1595 | (MAX_ACCEPT_TIMEOUT * 1000 \ | |
1596 | - ((MAX_ACCEPT_TIMEOUT - MIN_ACCEPT_TIMEOUT) * 1000 * nused) / (2 * nconns)) | |
1597 | ||
1598 | int n = poll (conns, nused, MAIN_THREAD_TIMEOUT); | |
1599 | ||
1600 | time_t now = time (NULL); | |
1601 | ||
1602 | /* If there is a descriptor ready for reading or there is a new | |
1603 | connection, process this now. */ | |
1604 | if (n > 0) | |
67479a70 | 1605 | { |
1945c96f UD |
1606 | if (conns[0].revents != 0) |
1607 | { | |
1608 | /* We have a new incoming connection. Accept the connection. */ | |
1609 | int fd = TEMP_FAILURE_RETRY (accept (sock, NULL, NULL)); | |
1610 | ||
9388dcbb UD |
1611 | /* Use the descriptor if we have not reached the limit. */ |
1612 | if (fd >= 0) | |
1945c96f | 1613 | { |
9388dcbb UD |
1614 | if (firstfree < nconns) |
1615 | { | |
1616 | conns[firstfree].fd = fd; | |
1617 | conns[firstfree].events = POLLRDNORM; | |
1618 | starttime[firstfree] = now; | |
1619 | if (firstfree >= nused) | |
1620 | nused = firstfree + 1; | |
1621 | ||
1622 | do | |
1623 | ++firstfree; | |
1624 | while (firstfree < nused && conns[firstfree].fd != -1); | |
1625 | } | |
1626 | else | |
1627 | /* We cannot use the connection so close it. */ | |
1628 | close (fd); | |
1945c96f UD |
1629 | } |
1630 | ||
1945c96f UD |
1631 | --n; |
1632 | } | |
1633 | ||
1634 | for (size_t cnt = 1; cnt < nused && n > 0; ++cnt) | |
1635 | if (conns[cnt].revents != 0) | |
1636 | { | |
fc03df7a | 1637 | fd_ready (conns[cnt].fd); |
1945c96f UD |
1638 | |
1639 | /* Clean up the CONNS array. */ | |
1640 | conns[cnt].fd = -1; | |
1641 | if (cnt < firstfree) | |
1642 | firstfree = cnt; | |
1643 | if (cnt == nused - 1) | |
1644 | do | |
1645 | --nused; | |
1646 | while (conns[nused - 1].fd == -1); | |
1647 | ||
1648 | --n; | |
1649 | } | |
1650 | } | |
1651 | ||
1652 | /* Now find entries which have timed out. */ | |
1653 | assert (nused > 0); | |
fc03df7a UD |
1654 | |
1655 | /* We make the timeout length depend on the number of file | |
1656 | descriptors currently used. */ | |
1945c96f UD |
1657 | #define ACCEPT_TIMEOUT \ |
1658 | (MAX_ACCEPT_TIMEOUT \ | |
1659 | - ((MAX_ACCEPT_TIMEOUT - MIN_ACCEPT_TIMEOUT) * nused) / nconns) | |
fc03df7a | 1660 | time_t laststart = now - ACCEPT_TIMEOUT; |
1945c96f | 1661 | |
fc03df7a UD |
1662 | for (size_t cnt = nused - 1; cnt > 0; --cnt) |
1663 | { | |
1945c96f UD |
1664 | if (conns[cnt].fd != -1 && starttime[cnt] < laststart) |
1665 | { | |
1666 | /* Remove the entry, it timed out. */ | |
1667 | (void) close (conns[cnt].fd); | |
1668 | conns[cnt].fd = -1; | |
1669 | ||
1670 | if (cnt < firstfree) | |
1671 | firstfree = cnt; | |
1672 | if (cnt == nused - 1) | |
1673 | do | |
1674 | --nused; | |
1675 | while (conns[nused - 1].fd == -1); | |
1676 | } | |
67479a70 | 1677 | } |
4401d759 UD |
1678 | |
1679 | if (restart_p (now)) | |
1680 | restart (); | |
67308730 | 1681 | } |
d67281a7 UD |
1682 | } |
1683 | ||
67479a70 | 1684 | |
fc03df7a UD |
1685 | #ifdef HAVE_EPOLL |
1686 | static void | |
1687 | main_loop_epoll (int efd) | |
1688 | { | |
1689 | struct epoll_event ev = { 0, }; | |
1690 | int nused = 1; | |
1691 | size_t highest = 0; | |
1692 | ||
1693 | /* Add the socket. */ | |
1694 | ev.events = EPOLLRDNORM; | |
1695 | ev.data.fd = sock; | |
1696 | if (epoll_ctl (efd, EPOLL_CTL_ADD, sock, &ev) == -1) | |
1697 | /* We cannot use epoll. */ | |
1698 | return; | |
1699 | ||
1700 | while (1) | |
1701 | { | |
1702 | struct epoll_event revs[100]; | |
1703 | # define nrevs (sizeof (revs) / sizeof (revs[0])) | |
1704 | ||
1705 | int n = epoll_wait (efd, revs, nrevs, MAIN_THREAD_TIMEOUT); | |
1706 | ||
1707 | time_t now = time (NULL); | |
1708 | ||
1709 | for (int cnt = 0; cnt < n; ++cnt) | |
1710 | if (revs[cnt].data.fd == sock) | |
1711 | { | |
1712 | /* A new connection. */ | |
1713 | int fd = TEMP_FAILURE_RETRY (accept (sock, NULL, NULL)); | |
1714 | ||
1715 | if (fd >= 0) | |
1716 | { | |
1717 | /* Try to add the new descriptor. */ | |
1718 | ev.data.fd = fd; | |
1719 | if (fd >= nconns | |
1720 | || epoll_ctl (efd, EPOLL_CTL_ADD, fd, &ev) == -1) | |
1721 | /* The descriptor is too large or something went | |
1722 | wrong. Close the descriptor. */ | |
1723 | close (fd); | |
1724 | else | |
1725 | { | |
1726 | /* Remember when we accepted the connection. */ | |
1727 | starttime[fd] = now; | |
1728 | ||
1729 | if (fd > highest) | |
1730 | highest = fd; | |
1731 | ||
1732 | ++nused; | |
1733 | } | |
1734 | } | |
1735 | } | |
1736 | else | |
1737 | { | |
1738 | /* Remove the descriptor from the epoll descriptor. */ | |
908c9e87 | 1739 | (void) epoll_ctl (efd, EPOLL_CTL_DEL, revs[cnt].data.fd, NULL); |
fc03df7a | 1740 | |
2461e3dc | 1741 | /* Get a worker to handle the request. */ |
fc03df7a UD |
1742 | fd_ready (revs[cnt].data.fd); |
1743 | ||
1744 | /* Reset the time. */ | |
1745 | starttime[revs[cnt].data.fd] = 0; | |
1746 | if (revs[cnt].data.fd == highest) | |
1747 | do | |
1748 | --highest; | |
1749 | while (highest > 0 && starttime[highest] == 0); | |
1750 | ||
1751 | --nused; | |
1752 | } | |
1753 | ||
1754 | /* Now look for descriptors for accepted connections which have | |
1755 | no reply in too long of a time. */ | |
1756 | time_t laststart = now - ACCEPT_TIMEOUT; | |
1757 | for (int cnt = highest; cnt > STDERR_FILENO; --cnt) | |
1758 | if (cnt != sock && starttime[cnt] != 0 && starttime[cnt] < laststart) | |
1759 | { | |
1760 | /* We are waiting for this one for too long. Close it. */ | |
908c9e87 | 1761 | (void) epoll_ctl (efd, EPOLL_CTL_DEL, cnt, NULL); |
fc03df7a UD |
1762 | |
1763 | (void) close (cnt); | |
1764 | ||
1765 | starttime[cnt] = 0; | |
1766 | if (cnt == highest) | |
1767 | --highest; | |
1768 | } | |
1769 | else if (cnt != sock && starttime[cnt] == 0 && cnt == highest) | |
1770 | --highest; | |
4401d759 UD |
1771 | |
1772 | if (restart_p (now)) | |
1773 | restart (); | |
fc03df7a UD |
1774 | } |
1775 | } | |
1776 | #endif | |
1777 | ||
1778 | ||
67479a70 | 1779 | /* Start all the threads we want. The initial process is thread no. 1. */ |
d67281a7 | 1780 | void |
67479a70 | 1781 | start_threads (void) |
d67281a7 | 1782 | { |
1945c96f UD |
1783 | /* Initialize the conditional variable we will use. The only |
1784 | non-standard attribute we might use is the clock selection. */ | |
1785 | pthread_condattr_t condattr; | |
1786 | pthread_condattr_init (&condattr); | |
1787 | ||
3078cba2 UD |
1788 | #if defined _POSIX_CLOCK_SELECTION && _POSIX_CLOCK_SELECTION >= 0 \ |
1789 | && defined _POSIX_MONOTONIC_CLOCK && _POSIX_MONOTONIC_CLOCK >= 0 | |
1945c96f UD |
1790 | /* Determine whether the monotonous clock is available. */ |
1791 | struct timespec dummy; | |
3078cba2 | 1792 | # if _POSIX_MONOTONIC_CLOCK == 0 |
94d824f9 | 1793 | if (sysconf (_SC_MONOTONIC_CLOCK) > 0) |
3078cba2 UD |
1794 | # endif |
1795 | # if _POSIX_CLOCK_SELECTION == 0 | |
94d824f9 | 1796 | if (sysconf (_SC_CLOCK_SELECTION) > 0) |
3078cba2 | 1797 | # endif |
94d824f9 UD |
1798 | if (clock_getres (CLOCK_MONOTONIC, &dummy) == 0 |
1799 | && pthread_condattr_setclock (&condattr, CLOCK_MONOTONIC) == 0) | |
1800 | timeout_clock = CLOCK_MONOTONIC; | |
1945c96f | 1801 | #endif |
d67281a7 | 1802 | |
1945c96f UD |
1803 | pthread_cond_init (&readylist_cond, &condattr); |
1804 | pthread_condattr_destroy (&condattr); | |
1805 | ||
1806 | ||
1807 | /* Create the attribute for the threads. They are all created | |
1808 | detached. */ | |
67479a70 UD |
1809 | pthread_attr_init (&attr); |
1810 | pthread_attr_setdetachstate (&attr, PTHREAD_CREATE_DETACHED); | |
27e82856 UD |
1811 | /* Use 1MB stacks, twice as much for 64-bit architectures. */ |
1812 | pthread_attr_setstacksize (&attr, 1024 * 1024 * (sizeof (void *) / 4)); | |
d67281a7 | 1813 | |
67479a70 UD |
1814 | /* We allow less than LASTDB threads only for debugging. */ |
1815 | if (debug_level == 0) | |
1816 | nthreads = MAX (nthreads, lastdb); | |
d67281a7 | 1817 | |
27e82856 | 1818 | int nfailed = 0; |
1945c96f UD |
1819 | for (long int i = 0; i < nthreads; ++i) |
1820 | { | |
1821 | pthread_t th; | |
27e82856 UD |
1822 | if (pthread_create (&th, &attr, nscd_run, (void *) (i - nfailed)) != 0) |
1823 | ++nfailed; | |
1824 | } | |
1825 | if (nthreads - nfailed < lastdb) | |
1826 | { | |
1827 | /* We could not start enough threads. */ | |
1828 | dbg_log (_("could only start %d threads; terminating"), | |
1829 | nthreads - nfailed); | |
1830 | exit (1); | |
1945c96f | 1831 | } |
d67281a7 | 1832 | |
fc03df7a UD |
1833 | /* Determine how much room for descriptors we should initially |
1834 | allocate. This might need to change later if we cap the number | |
1835 | with MAXCONN. */ | |
1836 | const long int nfds = sysconf (_SC_OPEN_MAX); | |
1837 | #define MINCONN 32 | |
1838 | #define MAXCONN 16384 | |
1839 | if (nfds == -1 || nfds > MAXCONN) | |
1840 | nconns = MAXCONN; | |
1841 | else if (nfds < MINCONN) | |
1842 | nconns = MINCONN; | |
1843 | else | |
1844 | nconns = nfds; | |
1845 | ||
1846 | /* We need memory to pass descriptors on to the worker threads. */ | |
1847 | fdlist = (struct fdlist *) xcalloc (nconns, sizeof (fdlist[0])); | |
1848 | /* Array to keep track when connection was accepted. */ | |
1849 | starttime = (time_t *) xcalloc (nconns, sizeof (starttime[0])); | |
1850 | ||
1945c96f UD |
1851 | /* In the main thread we execute the loop which handles incoming |
1852 | connections. */ | |
fc03df7a UD |
1853 | #ifdef HAVE_EPOLL |
1854 | int efd = epoll_create (100); | |
1855 | if (efd != -1) | |
1856 | { | |
1857 | main_loop_epoll (efd); | |
1858 | close (efd); | |
1859 | } | |
1860 | #endif | |
1861 | ||
1862 | main_loop_poll (); | |
d67281a7 | 1863 | } |
057685e4 UD |
1864 | |
1865 | ||
1866 | /* Look up the uid, gid, and supplementary groups to run nscd as. When | |
1867 | this function is called, we are not listening on the nscd socket yet so | |
1868 | we can just use the ordinary lookup functions without causing a lockup */ | |
1869 | static void | |
1870 | begin_drop_privileges (void) | |
1871 | { | |
a95a08b4 | 1872 | struct passwd *pwd = getpwnam (server_user); |
057685e4 UD |
1873 | |
1874 | if (pwd == NULL) | |
1875 | { | |
1876 | dbg_log (_("Failed to run nscd as user '%s'"), server_user); | |
1877 | error (EXIT_FAILURE, 0, _("Failed to run nscd as user '%s'"), | |
1878 | server_user); | |
1879 | } | |
1880 | ||
1881 | server_uid = pwd->pw_uid; | |
1882 | server_gid = pwd->pw_gid; | |
1883 | ||
4401d759 UD |
1884 | /* Save the old UID/GID if we have to change back. */ |
1885 | if (paranoia) | |
1886 | { | |
1887 | old_uid = getuid (); | |
1888 | old_gid = getgid (); | |
1889 | } | |
1890 | ||
a95a08b4 UD |
1891 | if (getgrouplist (server_user, server_gid, NULL, &server_ngroups) == 0) |
1892 | { | |
1893 | /* This really must never happen. */ | |
1894 | dbg_log (_("Failed to run nscd as user '%s'"), server_user); | |
1895 | error (EXIT_FAILURE, errno, _("initial getgrouplist failed")); | |
1896 | } | |
057685e4 | 1897 | |
a95a08b4 | 1898 | server_groups = (gid_t *) xmalloc (server_ngroups * sizeof (gid_t)); |
057685e4 UD |
1899 | |
1900 | if (getgrouplist (server_user, server_gid, server_groups, &server_ngroups) | |
1901 | == -1) | |
1902 | { | |
1903 | dbg_log (_("Failed to run nscd as user '%s'"), server_user); | |
1904 | error (EXIT_FAILURE, errno, _("getgrouplist failed")); | |
1905 | } | |
1906 | } | |
1907 | ||
1908 | ||
1909 | /* Call setgroups(), setgid(), and setuid() to drop root privileges and | |
1910 | run nscd as the user specified in the configuration file. */ | |
1911 | static void | |
1912 | finish_drop_privileges (void) | |
1913 | { | |
1f063dca UD |
1914 | #if defined HAVE_LIBAUDIT && defined HAVE_LIBCAP |
1915 | /* We need to preserve the capabilities to connect to the audit daemon. */ | |
1916 | cap_t new_caps = preserve_capabilities (); | |
1917 | #endif | |
1918 | ||
057685e4 UD |
1919 | if (setgroups (server_ngroups, server_groups) == -1) |
1920 | { | |
1921 | dbg_log (_("Failed to run nscd as user '%s'"), server_user); | |
1922 | error (EXIT_FAILURE, errno, _("setgroups failed")); | |
1923 | } | |
1924 | ||
91287339 UD |
1925 | int res; |
1926 | if (paranoia) | |
1927 | res = setresgid (server_gid, server_gid, old_gid); | |
1928 | else | |
1929 | res = setgid (server_gid); | |
1930 | if (res == -1) | |
057685e4 UD |
1931 | { |
1932 | dbg_log (_("Failed to run nscd as user '%s'"), server_user); | |
1933 | perror ("setgid"); | |
64d64de6 | 1934 | exit (4); |
057685e4 UD |
1935 | } |
1936 | ||
91287339 UD |
1937 | if (paranoia) |
1938 | res = setresuid (server_uid, server_uid, old_uid); | |
1939 | else | |
1940 | res = setuid (server_uid); | |
1941 | if (res == -1) | |
057685e4 UD |
1942 | { |
1943 | dbg_log (_("Failed to run nscd as user '%s'"), server_user); | |
1944 | perror ("setuid"); | |
64d64de6 | 1945 | exit (4); |
057685e4 | 1946 | } |
1f063dca UD |
1947 | |
1948 | #if defined HAVE_LIBAUDIT && defined HAVE_LIBCAP | |
1949 | /* Remove the temporary capabilities. */ | |
1950 | install_real_capabilities (new_caps); | |
1951 | #endif | |
057685e4 | 1952 | } |