]>
Commit | Line | Data |
---|---|---|
67479a70 | 1 | /* Inner loops of cache daemon. |
6cac6927 | 2 | Copyright (C) 1998-2003, 2004, 2005, 2006 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 UD |
77 | /* Map request type to a string. */ |
78 | const char *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 UD |
94 | [GETAI] = "GETAI", |
95 | [INITGROUPS] = "INITGROUPS" | |
67479a70 UD |
96 | }; |
97 | ||
98 | /* The control data structures for the services. */ | |
a95a08b4 | 99 | struct database_dyn dbs[lastdb] = |
67479a70 UD |
100 | { |
101 | [pwddb] = { | |
c2e13112 RM |
102 | .lock = PTHREAD_RWLOCK_WRITER_NONRECURSIVE_INITIALIZER_NP, |
103 | .enabled = 0, | |
104 | .check_file = 1, | |
a95a08b4 | 105 | .persistent = 0, |
797ed6f7 | 106 | .propagate = 1, |
c207f23b | 107 | .shared = 0, |
2c210d1e | 108 | .max_db_size = DEFAULT_MAX_DB_SIZE, |
c2e13112 | 109 | .filename = "/etc/passwd", |
a95a08b4 | 110 | .db_filename = _PATH_NSCD_PASSWD_DB, |
c2e13112 RM |
111 | .disabled_iov = &pwd_iov_disabled, |
112 | .postimeout = 3600, | |
a95a08b4 UD |
113 | .negtimeout = 20, |
114 | .wr_fd = -1, | |
115 | .ro_fd = -1, | |
116 | .mmap_used = false | |
67479a70 UD |
117 | }, |
118 | [grpdb] = { | |
c2e13112 RM |
119 | .lock = PTHREAD_RWLOCK_WRITER_NONRECURSIVE_INITIALIZER_NP, |
120 | .enabled = 0, | |
121 | .check_file = 1, | |
a95a08b4 | 122 | .persistent = 0, |
797ed6f7 | 123 | .propagate = 1, |
c207f23b | 124 | .shared = 0, |
2c210d1e | 125 | .max_db_size = DEFAULT_MAX_DB_SIZE, |
c2e13112 | 126 | .filename = "/etc/group", |
a95a08b4 | 127 | .db_filename = _PATH_NSCD_GROUP_DB, |
c2e13112 RM |
128 | .disabled_iov = &grp_iov_disabled, |
129 | .postimeout = 3600, | |
a95a08b4 UD |
130 | .negtimeout = 60, |
131 | .wr_fd = -1, | |
132 | .ro_fd = -1, | |
133 | .mmap_used = false | |
67479a70 UD |
134 | }, |
135 | [hstdb] = { | |
c2e13112 RM |
136 | .lock = PTHREAD_RWLOCK_WRITER_NONRECURSIVE_INITIALIZER_NP, |
137 | .enabled = 0, | |
138 | .check_file = 1, | |
a95a08b4 | 139 | .persistent = 0, |
797ed6f7 | 140 | .propagate = 0, /* Not used. */ |
c207f23b | 141 | .shared = 0, |
2c210d1e | 142 | .max_db_size = DEFAULT_MAX_DB_SIZE, |
c2e13112 | 143 | .filename = "/etc/hosts", |
a95a08b4 | 144 | .db_filename = _PATH_NSCD_HOSTS_DB, |
c2e13112 RM |
145 | .disabled_iov = &hst_iov_disabled, |
146 | .postimeout = 3600, | |
a95a08b4 UD |
147 | .negtimeout = 20, |
148 | .wr_fd = -1, | |
149 | .ro_fd = -1, | |
150 | .mmap_used = false | |
67479a70 UD |
151 | } |
152 | }; | |
d67281a7 | 153 | |
a95a08b4 UD |
154 | |
155 | /* Mapping of request type to database. */ | |
c207f23b | 156 | static struct database_dyn *const serv2db[LASTREQ] = |
a95a08b4 UD |
157 | { |
158 | [GETPWBYNAME] = &dbs[pwddb], | |
159 | [GETPWBYUID] = &dbs[pwddb], | |
160 | [GETGRBYNAME] = &dbs[grpdb], | |
161 | [GETGRBYGID] = &dbs[grpdb], | |
162 | [GETHOSTBYNAME] = &dbs[hstdb], | |
163 | [GETHOSTBYNAMEv6] = &dbs[hstdb], | |
164 | [GETHOSTBYADDR] = &dbs[hstdb], | |
c207f23b UD |
165 | [GETHOSTBYADDRv6] = &dbs[hstdb], |
166 | [GETFDPW] = &dbs[pwddb], | |
167 | [GETFDGR] = &dbs[grpdb], | |
168 | [GETFDHST] = &dbs[hstdb], | |
d19687d6 | 169 | [GETAI] = &dbs[hstdb], |
f7e7a396 | 170 | [INITGROUPS] = &dbs[grpdb] |
a95a08b4 UD |
171 | }; |
172 | ||
173 | ||
34489d95 UD |
174 | /* Number of seconds between two cache pruning runs. */ |
175 | #define CACHE_PRUNE_INTERVAL 15 | |
176 | ||
a95a08b4 | 177 | |
27e82856 | 178 | /* Initial number of threads to use. */ |
67479a70 | 179 | int nthreads = -1; |
27e82856 UD |
180 | /* Maximum number of threads to use. */ |
181 | int max_nthreads = 32; | |
d67281a7 | 182 | |
67479a70 UD |
183 | /* Socket for incoming connections. */ |
184 | static int sock; | |
d67281a7 | 185 | |
0fdb4f42 UD |
186 | /* Number of times clients had to wait. */ |
187 | unsigned long int client_queued; | |
188 | ||
d67281a7 | 189 | |
d2dc7d84 UD |
190 | ssize_t |
191 | writeall (int fd, const void *buf, size_t len) | |
192 | { | |
193 | size_t n = len; | |
194 | ssize_t ret; | |
195 | do | |
196 | { | |
2c210d1e | 197 | ret = TEMP_FAILURE_RETRY (send (fd, buf, n, MSG_NOSIGNAL)); |
d2dc7d84 UD |
198 | if (ret <= 0) |
199 | break; | |
200 | buf = (const char *) buf + ret; | |
201 | n -= ret; | |
202 | } | |
203 | while (n > 0); | |
204 | return ret < 0 ? ret : len - n; | |
205 | } | |
206 | ||
207 | ||
bd547139 UD |
208 | #ifdef HAVE_SENDFILE |
209 | ssize_t | |
210 | sendfileall (int tofd, int fromfd, off_t off, size_t len) | |
211 | { | |
212 | ssize_t n = len; | |
213 | ssize_t ret; | |
214 | ||
215 | do | |
216 | { | |
217 | ret = TEMP_FAILURE_RETRY (sendfile (tofd, fromfd, &off, n)); | |
218 | if (ret <= 0) | |
219 | break; | |
220 | n -= ret; | |
221 | } | |
222 | while (n > 0); | |
223 | return ret < 0 ? ret : len - n; | |
224 | } | |
225 | #endif | |
226 | ||
227 | ||
dc4bb1c2 UD |
228 | enum usekey |
229 | { | |
230 | use_not = 0, | |
231 | /* The following three are not really used, they are symbolic constants. */ | |
232 | use_first = 16, | |
233 | use_begin = 32, | |
234 | use_end = 64, | |
235 | ||
236 | use_he = 1, | |
237 | use_he_begin = use_he | use_begin, | |
238 | use_he_end = use_he | use_end, | |
239 | #if SEPARATE_KEY | |
240 | use_key = 2, | |
241 | use_key_begin = use_key | use_begin, | |
242 | use_key_end = use_key | use_end, | |
243 | use_key_first = use_key_begin | use_first, | |
244 | #endif | |
245 | use_data = 3, | |
246 | use_data_begin = use_data | use_begin, | |
247 | use_data_end = use_data | use_end, | |
248 | use_data_first = use_data_begin | use_first | |
249 | }; | |
250 | ||
251 | ||
252 | static int | |
253 | check_use (const char *data, nscd_ssize_t first_free, uint8_t *usemap, | |
254 | enum usekey use, ref_t start, size_t len) | |
255 | { | |
256 | assert (len >= 2); | |
257 | ||
258 | if (start > first_free || start + len > first_free | |
259 | || (start & BLOCK_ALIGN_M1)) | |
260 | return 0; | |
261 | ||
262 | if (usemap[start] == use_not) | |
263 | { | |
264 | /* Add the start marker. */ | |
265 | usemap[start] = use | use_begin; | |
266 | use &= ~use_first; | |
267 | ||
268 | while (--len > 0) | |
269 | if (usemap[++start] != use_not) | |
270 | return 0; | |
271 | else | |
272 | usemap[start] = use; | |
273 | ||
274 | /* Add the end marker. */ | |
275 | usemap[start] = use | use_end; | |
276 | } | |
277 | else if ((usemap[start] & ~use_first) == ((use | use_begin) & ~use_first)) | |
278 | { | |
279 | /* Hash entries can't be shared. */ | |
280 | if (use == use_he) | |
281 | return 0; | |
282 | ||
283 | usemap[start] |= (use & use_first); | |
284 | use &= ~use_first; | |
285 | ||
286 | while (--len > 1) | |
287 | if (usemap[++start] != use) | |
288 | return 0; | |
289 | ||
290 | if (usemap[++start] != (use | use_end)) | |
291 | return 0; | |
292 | } | |
293 | else | |
294 | /* Points to a wrong object or somewhere in the middle. */ | |
295 | return 0; | |
296 | ||
297 | return 1; | |
298 | } | |
299 | ||
300 | ||
301 | /* Verify data in persistent database. */ | |
302 | static int | |
303 | verify_persistent_db (void *mem, struct database_pers_head *readhead, int dbnr) | |
304 | { | |
305 | assert (dbnr == pwddb || dbnr == grpdb || dbnr == hstdb); | |
306 | ||
307 | time_t now = time (NULL); | |
308 | ||
309 | struct database_pers_head *head = mem; | |
310 | struct database_pers_head head_copy = *head; | |
311 | ||
312 | /* Check that the header that was read matches the head in the database. */ | |
313 | if (readhead != NULL && memcmp (head, readhead, sizeof (*head)) != 0) | |
314 | return 0; | |
315 | ||
316 | /* First some easy tests: make sure the database header is sane. */ | |
317 | if (head->version != DB_VERSION | |
318 | || head->header_size != sizeof (*head) | |
319 | /* We allow a timestamp to be one hour ahead of the current time. | |
320 | This should cover daylight saving time changes. */ | |
321 | || head->timestamp > now + 60 * 60 + 60 | |
322 | || (head->gc_cycle & 1) | |
323 | || (size_t) head->module > INT32_MAX / sizeof (ref_t) | |
324 | || (size_t) head->data_size > INT32_MAX - head->module * sizeof (ref_t) | |
325 | || head->first_free < 0 | |
326 | || head->first_free > head->data_size | |
327 | || (head->first_free & BLOCK_ALIGN_M1) != 0 | |
328 | || head->maxnentries < 0 | |
329 | || head->maxnsearched < 0) | |
330 | return 0; | |
331 | ||
332 | uint8_t *usemap = calloc (head->first_free, 1); | |
333 | if (usemap == NULL) | |
334 | return 0; | |
335 | ||
336 | const char *data = (char *) &head->array[roundup (head->module, | |
337 | ALIGN / sizeof (ref_t))]; | |
338 | ||
339 | nscd_ssize_t he_cnt = 0; | |
340 | for (nscd_ssize_t cnt = 0; cnt < head->module; ++cnt) | |
341 | { | |
342 | ref_t work = head->array[cnt]; | |
343 | ||
344 | while (work != ENDREF) | |
345 | { | |
346 | if (! check_use (data, head->first_free, usemap, use_he, work, | |
347 | sizeof (struct hashentry))) | |
348 | goto fail; | |
349 | ||
350 | /* Now we know we can dereference the record. */ | |
351 | struct hashentry *here = (struct hashentry *) (data + work); | |
352 | ||
353 | ++he_cnt; | |
354 | ||
355 | /* Make sure the record is for this type of service. */ | |
356 | if (here->type >= LASTREQ | |
357 | || serv2db[here->type] != &dbs[dbnr]) | |
358 | goto fail; | |
359 | ||
360 | /* Validate boolean field value. */ | |
361 | if (here->first != false && here->first != true) | |
362 | goto fail; | |
363 | ||
364 | if (here->len < 0) | |
365 | goto fail; | |
366 | ||
367 | /* Now the data. */ | |
368 | if (here->packet < 0 | |
369 | || here->packet > head->first_free | |
370 | || here->packet + sizeof (struct datahead) > head->first_free) | |
371 | goto fail; | |
372 | ||
373 | struct datahead *dh = (struct datahead *) (data + here->packet); | |
374 | ||
375 | if (! check_use (data, head->first_free, usemap, | |
376 | use_data | (here->first ? use_first : 0), | |
377 | here->packet, dh->allocsize)) | |
378 | goto fail; | |
379 | ||
380 | if (dh->allocsize < sizeof (struct datahead) | |
381 | || dh->recsize > dh->allocsize | |
382 | || (dh->notfound != false && dh->notfound != true) | |
383 | || (dh->usable != false && dh->usable != true)) | |
384 | goto fail; | |
385 | ||
386 | if (here->key < here->packet + sizeof (struct datahead) | |
387 | || here->key > here->packet + dh->allocsize | |
388 | || here->key + here->len > here->packet + dh->allocsize) | |
389 | { | |
390 | #if SEPARATE_KEY | |
391 | /* If keys can appear outside of data, this should be done | |
392 | instead. But gc doesn't mark the data in that case. */ | |
393 | if (! check_use (data, head->first_free, usemap, | |
394 | use_key | (here->first ? use_first : 0), | |
395 | here->key, here->len)) | |
396 | #endif | |
397 | goto fail; | |
398 | } | |
399 | ||
400 | work = here->next; | |
401 | } | |
402 | } | |
403 | ||
404 | if (he_cnt != head->nentries) | |
405 | goto fail; | |
406 | ||
407 | /* See if all data and keys had at least one reference from | |
408 | he->first == true hashentry. */ | |
409 | for (ref_t idx = 0; idx < head->first_free; ++idx) | |
410 | { | |
411 | #if SEPARATE_KEY | |
412 | if (usemap[idx] == use_key_begin) | |
413 | goto fail; | |
414 | #endif | |
415 | if (usemap[idx] == use_data_begin) | |
416 | goto fail; | |
417 | } | |
418 | ||
419 | /* Finally, make sure the database hasn't changed since the first test. */ | |
420 | if (memcmp (mem, &head_copy, sizeof (*head)) != 0) | |
421 | goto fail; | |
422 | ||
423 | free (usemap); | |
424 | return 1; | |
425 | ||
426 | fail: | |
427 | free (usemap); | |
428 | return 0; | |
429 | } | |
430 | ||
431 | ||
a334319f UD |
432 | /* Initialize database information structures. */ |
433 | void | |
434 | nscd_init (void) | |
0ecb606c | 435 | { |
057685e4 UD |
436 | /* Look up unprivileged uid/gid/groups before we start listening on the |
437 | socket */ | |
438 | if (server_user != NULL) | |
439 | begin_drop_privileges (); | |
440 | ||
67479a70 UD |
441 | if (nthreads == -1) |
442 | /* No configuration for this value, assume a default. */ | |
443 | nthreads = 2 * lastdb; | |
d67281a7 | 444 | |
d2dc7d84 | 445 | for (size_t cnt = 0; cnt < lastdb; ++cnt) |
67479a70 | 446 | if (dbs[cnt].enabled) |
9db29cde | 447 | { |
67479a70 | 448 | pthread_rwlock_init (&dbs[cnt].lock, NULL); |
a95a08b4 | 449 | pthread_mutex_init (&dbs[cnt].memlock, NULL); |
264d5b94 | 450 | |
a95a08b4 | 451 | if (dbs[cnt].persistent) |
e09edf23 | 452 | { |
a95a08b4 UD |
453 | /* Try to open the appropriate file on disk. */ |
454 | int fd = open (dbs[cnt].db_filename, O_RDWR); | |
455 | if (fd != -1) | |
456 | { | |
457 | struct stat64 st; | |
458 | void *mem; | |
459 | size_t total; | |
460 | struct database_pers_head head; | |
461 | ssize_t n = TEMP_FAILURE_RETRY (read (fd, &head, | |
462 | sizeof (head))); | |
463 | if (n != sizeof (head) || fstat64 (fd, &st) != 0) | |
464 | { | |
465 | fail_db: | |
466 | dbg_log (_("invalid persistent database file \"%s\": %s"), | |
467 | dbs[cnt].db_filename, strerror (errno)); | |
dc4bb1c2 | 468 | unlink (dbs[cnt].db_filename); |
a95a08b4 UD |
469 | } |
470 | else if (head.module == 0 && head.data_size == 0) | |
471 | { | |
472 | /* The file has been created, but the head has not been | |
473 | initialized yet. Remove the old file. */ | |
474 | unlink (dbs[cnt].db_filename); | |
475 | } | |
476 | else if (head.header_size != (int) sizeof (head)) | |
477 | { | |
478 | dbg_log (_("invalid persistent database file \"%s\": %s"), | |
479 | dbs[cnt].db_filename, | |
480 | _("header size does not match")); | |
dc4bb1c2 | 481 | unlink (dbs[cnt].db_filename); |
a95a08b4 UD |
482 | } |
483 | else if ((total = (sizeof (head) | |
c207f23b | 484 | + roundup (head.module * sizeof (ref_t), |
a95a08b4 UD |
485 | ALIGN) |
486 | + head.data_size)) | |
dc4bb1c2 UD |
487 | > st.st_size |
488 | || total < sizeof (head)) | |
a95a08b4 UD |
489 | { |
490 | dbg_log (_("invalid persistent database file \"%s\": %s"), | |
491 | dbs[cnt].db_filename, | |
492 | _("file size does not match")); | |
dc4bb1c2 | 493 | unlink (dbs[cnt].db_filename); |
a95a08b4 | 494 | } |
2c210d1e UD |
495 | /* Note we map with the maximum size allowed for the |
496 | database. This is likely much larger than the | |
497 | actual file size. This is OK on most OSes since | |
498 | extensions of the underlying file will | |
499 | automatically translate more pages available for | |
500 | memory access. */ | |
501 | else if ((mem = mmap (NULL, dbs[cnt].max_db_size, | |
502 | PROT_READ | PROT_WRITE, | |
503 | MAP_SHARED, fd, 0)) | |
504 | == MAP_FAILED) | |
a95a08b4 | 505 | goto fail_db; |
dc4bb1c2 UD |
506 | else if (!verify_persistent_db (mem, &head, cnt)) |
507 | { | |
508 | munmap (mem, total); | |
509 | dbg_log (_("invalid persistent database file \"%s\": %s"), | |
510 | dbs[cnt].db_filename, | |
511 | _("verification failed")); | |
512 | unlink (dbs[cnt].db_filename); | |
513 | } | |
a95a08b4 UD |
514 | else |
515 | { | |
516 | /* Success. We have the database. */ | |
517 | dbs[cnt].head = mem; | |
518 | dbs[cnt].memsize = total; | |
519 | dbs[cnt].data = (char *) | |
520 | &dbs[cnt].head->array[roundup (dbs[cnt].head->module, | |
521 | ALIGN / sizeof (ref_t))]; | |
522 | dbs[cnt].mmap_used = true; | |
523 | ||
524 | if (dbs[cnt].suggested_module > head.module) | |
525 | dbg_log (_("suggested size of table for database %s larger than the persistent database's table"), | |
526 | dbnames[cnt]); | |
527 | ||
528 | dbs[cnt].wr_fd = fd; | |
529 | fd = -1; | |
530 | /* We also need a read-only descriptor. */ | |
d13a3c57 UD |
531 | if (dbs[cnt].shared) |
532 | { | |
533 | dbs[cnt].ro_fd = open (dbs[cnt].db_filename, O_RDONLY); | |
534 | if (dbs[cnt].ro_fd == -1) | |
535 | dbg_log (_("\ | |
a95a08b4 | 536 | cannot create read-only descriptor for \"%s\"; no mmap"), |
d13a3c57 UD |
537 | dbs[cnt].db_filename); |
538 | } | |
a95a08b4 UD |
539 | |
540 | // XXX Shall we test whether the descriptors actually | |
541 | // XXX point to the same file? | |
542 | } | |
543 | ||
544 | /* Close the file descriptors in case something went | |
545 | wrong in which case the variable have not been | |
546 | assigned -1. */ | |
547 | if (fd != -1) | |
548 | close (fd); | |
549 | } | |
550 | } | |
551 | ||
552 | if (dbs[cnt].head == NULL) | |
553 | { | |
554 | /* No database loaded. Allocate the data structure, | |
555 | possibly on disk. */ | |
556 | struct database_pers_head head; | |
557 | size_t total = (sizeof (head) | |
558 | + roundup (dbs[cnt].suggested_module | |
559 | * sizeof (ref_t), ALIGN) | |
560 | + (dbs[cnt].suggested_module | |
561 | * DEFAULT_DATASIZE_PER_BUCKET)); | |
562 | ||
563 | /* Try to create the database. If we do not need a | |
564 | persistent database create a temporary file. */ | |
565 | int fd; | |
566 | int ro_fd = -1; | |
567 | if (dbs[cnt].persistent) | |
568 | { | |
569 | fd = open (dbs[cnt].db_filename, | |
570 | O_RDWR | O_CREAT | O_EXCL | O_TRUNC, | |
571 | S_IRUSR | S_IWUSR); | |
d13a3c57 | 572 | if (fd != -1 && dbs[cnt].shared) |
a95a08b4 UD |
573 | ro_fd = open (dbs[cnt].db_filename, O_RDONLY); |
574 | } | |
575 | else | |
576 | { | |
a8a58967 | 577 | char fname[] = _PATH_NSCD_XYZ_DB_TMP; |
a95a08b4 UD |
578 | fd = mkstemp (fname); |
579 | ||
580 | /* We do not need the file name anymore after we | |
581 | opened another file descriptor in read-only mode. */ | |
5ca3d19c | 582 | if (fd != -1) |
a95a08b4 | 583 | { |
5ca3d19c UD |
584 | if (dbs[cnt].shared) |
585 | ro_fd = open (fname, O_RDONLY); | |
a95a08b4 UD |
586 | |
587 | unlink (fname); | |
588 | } | |
589 | } | |
590 | ||
591 | if (fd == -1) | |
592 | { | |
593 | if (errno == EEXIST) | |
594 | { | |
595 | dbg_log (_("database for %s corrupted or simultaneously used; remove %s manually if necessary and restart"), | |
596 | dbnames[cnt], dbs[cnt].db_filename); | |
597 | // XXX Correct way to terminate? | |
598 | exit (1); | |
599 | } | |
600 | ||
601 | if (dbs[cnt].persistent) | |
602 | dbg_log (_("cannot create %s; no persistent database used"), | |
603 | dbs[cnt].db_filename); | |
604 | else | |
605 | dbg_log (_("cannot create %s; no sharing possible"), | |
606 | dbs[cnt].db_filename); | |
607 | ||
608 | dbs[cnt].persistent = 0; | |
609 | // XXX remember: no mmap | |
610 | } | |
611 | else | |
612 | { | |
613 | /* Tell the user if we could not create the read-only | |
614 | descriptor. */ | |
d13a3c57 | 615 | if (ro_fd == -1 && dbs[cnt].shared) |
a95a08b4 UD |
616 | dbg_log (_("\ |
617 | cannot create read-only descriptor for \"%s\"; no mmap"), | |
618 | dbs[cnt].db_filename); | |
619 | ||
620 | /* Before we create the header, initialiye the hash | |
621 | table. So that if we get interrupted if writing | |
622 | the header we can recognize a partially initialized | |
623 | database. */ | |
624 | size_t ps = sysconf (_SC_PAGESIZE); | |
625 | char tmpbuf[ps]; | |
626 | assert (~ENDREF == 0); | |
627 | memset (tmpbuf, '\xff', ps); | |
628 | ||
629 | size_t remaining = dbs[cnt].suggested_module * sizeof (ref_t); | |
630 | off_t offset = sizeof (head); | |
631 | ||
632 | size_t towrite; | |
633 | if (offset % ps != 0) | |
634 | { | |
635 | towrite = MIN (remaining, ps - (offset % ps)); | |
233399bc UD |
636 | if (pwrite (fd, tmpbuf, towrite, offset) != towrite) |
637 | goto write_fail; | |
a95a08b4 UD |
638 | offset += towrite; |
639 | remaining -= towrite; | |
640 | } | |
641 | ||
642 | while (remaining > ps) | |
643 | { | |
233399bc UD |
644 | if (pwrite (fd, tmpbuf, ps, offset) == -1) |
645 | goto write_fail; | |
a95a08b4 UD |
646 | offset += ps; |
647 | remaining -= ps; | |
648 | } | |
649 | ||
233399bc UD |
650 | if (remaining > 0 |
651 | && pwrite (fd, tmpbuf, remaining, offset) != remaining) | |
652 | goto write_fail; | |
a95a08b4 UD |
653 | |
654 | /* Create the header of the file. */ | |
655 | struct database_pers_head head = | |
656 | { | |
657 | .version = DB_VERSION, | |
658 | .header_size = sizeof (head), | |
659 | .module = dbs[cnt].suggested_module, | |
660 | .data_size = (dbs[cnt].suggested_module | |
661 | * DEFAULT_DATASIZE_PER_BUCKET), | |
662 | .first_free = 0 | |
663 | }; | |
664 | void *mem; | |
665 | ||
666 | if ((TEMP_FAILURE_RETRY (write (fd, &head, sizeof (head))) | |
667 | != sizeof (head)) | |
2c210d1e UD |
668 | || (TEMP_FAILURE_RETRY_VAL (posix_fallocate (fd, 0, total)) |
669 | != 0) | |
670 | || (mem = mmap (NULL, dbs[cnt].max_db_size, | |
671 | PROT_READ | PROT_WRITE, | |
a95a08b4 UD |
672 | MAP_SHARED, fd, 0)) == MAP_FAILED) |
673 | { | |
233399bc | 674 | write_fail: |
a95a08b4 UD |
675 | unlink (dbs[cnt].db_filename); |
676 | dbg_log (_("cannot write to database file %s: %s"), | |
677 | dbs[cnt].db_filename, strerror (errno)); | |
678 | dbs[cnt].persistent = 0; | |
679 | } | |
680 | else | |
681 | { | |
682 | /* Success. */ | |
683 | dbs[cnt].head = mem; | |
684 | dbs[cnt].data = (char *) | |
685 | &dbs[cnt].head->array[roundup (dbs[cnt].head->module, | |
686 | ALIGN / sizeof (ref_t))]; | |
687 | dbs[cnt].memsize = total; | |
688 | dbs[cnt].mmap_used = true; | |
689 | ||
690 | /* Remember the descriptors. */ | |
691 | dbs[cnt].wr_fd = fd; | |
692 | dbs[cnt].ro_fd = ro_fd; | |
693 | fd = -1; | |
694 | ro_fd = -1; | |
695 | } | |
696 | ||
697 | if (fd != -1) | |
698 | close (fd); | |
699 | if (ro_fd != -1) | |
700 | close (ro_fd); | |
701 | } | |
702 | } | |
703 | ||
4401d759 UD |
704 | if (paranoia |
705 | && ((dbs[cnt].wr_fd != -1 | |
706 | && fcntl (dbs[cnt].wr_fd, F_SETFD, FD_CLOEXEC) == -1) | |
707 | || (dbs[cnt].ro_fd != -1 | |
708 | && fcntl (dbs[cnt].ro_fd, F_SETFD, FD_CLOEXEC) == -1))) | |
709 | { | |
710 | dbg_log (_("\ | |
711 | cannot set socket to close on exec: %s; disabling paranoia mode"), | |
712 | strerror (errno)); | |
713 | paranoia = 0; | |
714 | } | |
715 | ||
a95a08b4 UD |
716 | if (dbs[cnt].head == NULL) |
717 | { | |
718 | /* We do not use the persistent database. Just | |
719 | create an in-memory data structure. */ | |
720 | assert (! dbs[cnt].persistent); | |
721 | ||
722 | dbs[cnt].head = xmalloc (sizeof (struct database_pers_head) | |
723 | + (dbs[cnt].suggested_module | |
724 | * sizeof (ref_t))); | |
a334319f | 725 | memset (dbs[cnt].head, '\0', sizeof (dbs[cnt].head)); |
a95a08b4 UD |
726 | assert (~ENDREF == 0); |
727 | memset (dbs[cnt].head->array, '\xff', | |
728 | dbs[cnt].suggested_module * sizeof (ref_t)); | |
729 | dbs[cnt].head->module = dbs[cnt].suggested_module; | |
730 | dbs[cnt].head->data_size = (DEFAULT_DATASIZE_PER_BUCKET | |
731 | * dbs[cnt].head->module); | |
732 | dbs[cnt].data = xmalloc (dbs[cnt].head->data_size); | |
733 | dbs[cnt].head->first_free = 0; | |
c207f23b UD |
734 | |
735 | dbs[cnt].shared = 0; | |
736 | assert (dbs[cnt].ro_fd == -1); | |
e09edf23 | 737 | } |
d67281a7 | 738 | |
67479a70 | 739 | if (dbs[cnt].check_file) |
d67281a7 | 740 | { |
67479a70 | 741 | /* We need the modification date of the file. */ |
80ea3037 | 742 | struct stat64 st; |
d67281a7 | 743 | |
80ea3037 | 744 | if (stat64 (dbs[cnt].filename, &st) < 0) |
d67281a7 | 745 | { |
67479a70 UD |
746 | /* We cannot stat() the file, disable file checking. */ |
747 | dbg_log (_("cannot stat() file `%s': %s"), | |
e09edf23 | 748 | dbs[cnt].filename, strerror (errno)); |
67479a70 | 749 | dbs[cnt].check_file = 0; |
d67281a7 UD |
750 | } |
751 | else | |
67479a70 UD |
752 | dbs[cnt].file_mtime = st.st_mtime; |
753 | } | |
754 | } | |
d67281a7 UD |
755 | |
756 | /* Create the socket. */ | |
67479a70 UD |
757 | sock = socket (AF_UNIX, SOCK_STREAM, 0); |
758 | if (sock < 0) | |
d67281a7 | 759 | { |
67479a70 | 760 | dbg_log (_("cannot open socket: %s"), strerror (errno)); |
64d64de6 | 761 | exit (errno == EACCES ? 4 : 1); |
d67281a7 UD |
762 | } |
763 | /* Bind a name to the socket. */ | |
d2dc7d84 | 764 | struct sockaddr_un sock_addr; |
d67281a7 UD |
765 | sock_addr.sun_family = AF_UNIX; |
766 | strcpy (sock_addr.sun_path, _PATH_NSCDSOCKET); | |
67479a70 | 767 | if (bind (sock, (struct sockaddr *) &sock_addr, sizeof (sock_addr)) < 0) |
d67281a7 UD |
768 | { |
769 | dbg_log ("%s: %s", _PATH_NSCDSOCKET, strerror (errno)); | |
64d64de6 | 770 | exit (errno == EACCES ? 4 : 1); |
d67281a7 | 771 | } |
67479a70 | 772 | |
4401d759 | 773 | /* We don't want to get stuck on accept. */ |
d6db0975 | 774 | int fl = fcntl (sock, F_GETFL); |
4401d759 UD |
775 | if (fl == -1 || fcntl (sock, F_SETFL, fl | O_NONBLOCK) == -1) |
776 | { | |
777 | dbg_log (_("cannot change socket to nonblocking mode: %s"), | |
778 | strerror (errno)); | |
779 | exit (1); | |
780 | } | |
781 | ||
782 | /* The descriptor needs to be closed on exec. */ | |
783 | if (paranoia && fcntl (sock, F_SETFD, FD_CLOEXEC) == -1) | |
784 | { | |
785 | dbg_log (_("cannot set socket to close on exec: %s"), | |
786 | strerror (errno)); | |
787 | exit (1); | |
788 | } | |
d6db0975 | 789 | |
d67281a7 | 790 | /* Set permissions for the socket. */ |
a95a08b4 | 791 | chmod (_PATH_NSCDSOCKET, DEFFILEMODE); |
d67281a7 UD |
792 | |
793 | /* Set the socket up to accept connections. */ | |
67479a70 | 794 | if (listen (sock, SOMAXCONN) < 0) |
d67281a7 | 795 | { |
67479a70 UD |
796 | dbg_log (_("cannot enable socket to accept connections: %s"), |
797 | strerror (errno)); | |
d67281a7 UD |
798 | exit (1); |
799 | } | |
057685e4 UD |
800 | |
801 | /* Change to unprivileged uid/gid/groups if specifed in config file */ | |
802 | if (server_user != NULL) | |
803 | finish_drop_privileges (); | |
d67281a7 UD |
804 | } |
805 | ||
67479a70 UD |
806 | |
807 | /* Close the connections. */ | |
d67281a7 | 808 | void |
67479a70 | 809 | close_sockets (void) |
d67281a7 | 810 | { |
67479a70 UD |
811 | close (sock); |
812 | } | |
d67281a7 | 813 | |
a12ce44f | 814 | |
756409c4 | 815 | static void |
a334319f | 816 | invalidate_cache (char *key) |
756409c4 UD |
817 | { |
818 | dbtype number; | |
819 | ||
820 | if (strcmp (key, "passwd") == 0) | |
821 | number = pwddb; | |
822 | else if (strcmp (key, "group") == 0) | |
823 | number = grpdb; | |
23700036 | 824 | else if (__builtin_expect (strcmp (key, "hosts"), 0) == 0) |
482bbeb9 UD |
825 | { |
826 | number = hstdb; | |
827 | ||
828 | /* Re-initialize the resolver. resolv.conf might have changed. */ | |
829 | res_init (); | |
830 | } | |
23700036 | 831 | else |
a334319f | 832 | return; |
756409c4 | 833 | |
fd665070 | 834 | if (dbs[number].enabled) |
a334319f | 835 | prune_cache (&dbs[number], LONG_MAX); |
756409c4 UD |
836 | } |
837 | ||
67479a70 | 838 | |
c207f23b UD |
839 | #ifdef SCM_RIGHTS |
840 | static void | |
841 | send_ro_fd (struct database_dyn *db, char *key, int fd) | |
842 | { | |
843 | /* If we do not have an read-only file descriptor do nothing. */ | |
844 | if (db->ro_fd == -1) | |
845 | return; | |
846 | ||
847 | /* We need to send some data along with the descriptor. */ | |
848 | struct iovec iov[1]; | |
849 | iov[0].iov_base = key; | |
850 | iov[0].iov_len = strlen (key) + 1; | |
851 | ||
852 | /* Prepare the control message to transfer the descriptor. */ | |
a08ab897 UD |
853 | union |
854 | { | |
855 | struct cmsghdr hdr; | |
856 | char bytes[CMSG_SPACE (sizeof (int))]; | |
857 | } buf; | |
c207f23b | 858 | struct msghdr msg = { .msg_iov = iov, .msg_iovlen = 1, |
a08ab897 UD |
859 | .msg_control = buf.bytes, |
860 | .msg_controllen = sizeof (buf) }; | |
c207f23b UD |
861 | struct cmsghdr *cmsg = CMSG_FIRSTHDR (&msg); |
862 | ||
863 | cmsg->cmsg_level = SOL_SOCKET; | |
864 | cmsg->cmsg_type = SCM_RIGHTS; | |
865 | cmsg->cmsg_len = CMSG_LEN (sizeof (int)); | |
866 | ||
867 | *(int *) CMSG_DATA (cmsg) = db->ro_fd; | |
868 | ||
869 | msg.msg_controllen = cmsg->cmsg_len; | |
870 | ||
871 | /* Send the control message. We repeat when we are interrupted but | |
872 | everything else is ignored. */ | |
6925ef9a UD |
873 | #ifndef MSG_NOSIGNAL |
874 | # define MSG_NOSIGNAL 0 | |
875 | #endif | |
876 | (void) TEMP_FAILURE_RETRY (sendmsg (fd, &msg, MSG_NOSIGNAL)); | |
c207f23b UD |
877 | |
878 | if (__builtin_expect (debug_level > 0, 0)) | |
879 | dbg_log (_("provide access to FD %d, for %s"), db->ro_fd, key); | |
880 | } | |
881 | #endif /* SCM_RIGHTS */ | |
882 | ||
883 | ||
67479a70 UD |
884 | /* Handle new request. */ |
885 | static void | |
a1c542bf | 886 | handle_request (int fd, request_header *req, void *key, uid_t uid) |
67479a70 | 887 | { |
23700036 | 888 | if (__builtin_expect (req->version, NSCD_VERSION) != NSCD_VERSION) |
d67281a7 | 889 | { |
98e75a1c UD |
890 | if (debug_level > 0) |
891 | dbg_log (_("\ | |
67479a70 | 892 | cannot handle old request version %d; current version is %d"), |
98e75a1c | 893 | req->version, NSCD_VERSION); |
d67281a7 UD |
894 | return; |
895 | } | |
896 | ||
74a30a58 UD |
897 | /* Make the SELinux check before we go on to the standard checks. We |
898 | need to verify that the request type is valid, since it has not | |
899 | yet been checked at this point. */ | |
900 | if (selinux_enabled | |
901 | && __builtin_expect (req->type, GETPWBYNAME) >= GETPWBYNAME | |
902 | && __builtin_expect (req->type, LASTREQ) < LASTREQ | |
903 | && nscd_request_avc_has_perm (fd, req->type) != 0) | |
904 | return; | |
905 | ||
a95a08b4 UD |
906 | struct database_dyn *db = serv2db[req->type]; |
907 | ||
f7e7a396 UD |
908 | // XXX Clean up so that each new command need not introduce a |
909 | // XXX new conditional. | |
d19687d6 UD |
910 | if ((__builtin_expect (req->type, GETPWBYNAME) >= GETPWBYNAME |
911 | && __builtin_expect (req->type, LASTDBREQ) <= LASTDBREQ) | |
f7e7a396 | 912 | || req->type == GETAI || req->type == INITGROUPS) |
d67281a7 | 913 | { |
23700036 | 914 | if (__builtin_expect (debug_level, 0) > 0) |
8d8c6efa UD |
915 | { |
916 | if (req->type == GETHOSTBYADDR || req->type == GETHOSTBYADDRv6) | |
917 | { | |
918 | char buf[INET6_ADDRSTRLEN]; | |
919 | ||
920 | dbg_log ("\t%s (%s)", serv2str[req->type], | |
921 | inet_ntop (req->type == GETHOSTBYADDR | |
922 | ? AF_INET : AF_INET6, | |
923 | key, buf, sizeof (buf))); | |
924 | } | |
925 | else | |
a95a08b4 | 926 | dbg_log ("\t%s (%s)", serv2str[req->type], (char *) key); |
8d8c6efa | 927 | } |
d67281a7 | 928 | |
67479a70 UD |
929 | /* Is this service enabled? */ |
930 | if (!db->enabled) | |
931 | { | |
ce85d65b | 932 | /* No, sent the prepared record. */ |
2c210d1e UD |
933 | if (TEMP_FAILURE_RETRY (send (fd, db->disabled_iov->iov_base, |
934 | db->disabled_iov->iov_len, | |
935 | MSG_NOSIGNAL)) | |
4c5dd2a2 | 936 | != (ssize_t) db->disabled_iov->iov_len |
23700036 | 937 | && __builtin_expect (debug_level, 0) > 0) |
67479a70 UD |
938 | { |
939 | /* We have problems sending the result. */ | |
940 | char buf[256]; | |
941 | dbg_log (_("cannot write result: %s"), | |
942 | strerror_r (errno, buf, sizeof (buf))); | |
943 | } | |
d67281a7 | 944 | |
67479a70 UD |
945 | return; |
946 | } | |
d67281a7 | 947 | |
67479a70 | 948 | /* Be sure we can read the data. */ |
c86e6aec UD |
949 | if (__builtin_expect (pthread_rwlock_tryrdlock (&db->lock) != 0, 0)) |
950 | { | |
a95a08b4 | 951 | ++db->head->rdlockdelayed; |
c86e6aec UD |
952 | pthread_rwlock_rdlock (&db->lock); |
953 | } | |
67479a70 UD |
954 | |
955 | /* See whether we can handle it from the cache. */ | |
a95a08b4 UD |
956 | struct datahead *cached; |
957 | cached = (struct datahead *) cache_search (req->type, key, req->key_len, | |
958 | db, uid); | |
67479a70 UD |
959 | if (cached != NULL) |
960 | { | |
961 | /* Hurray it's in the cache. */ | |
eac10791 UD |
962 | ssize_t nwritten; |
963 | ||
964 | #ifdef HAVE_SENDFILE | |
965 | if (db->mmap_used || !cached->notfound) | |
966 | { | |
967 | assert (db->wr_fd != -1); | |
968 | assert ((char *) cached->data > (char *) db->data); | |
969 | assert ((char *) cached->data - (char *) db->head | |
970 | + cached->recsize | |
971 | <= (sizeof (struct database_pers_head) | |
972 | + db->head->module * sizeof (ref_t) | |
973 | + db->head->data_size)); | |
bd547139 UD |
974 | nwritten = sendfileall (fd, db->wr_fd, |
975 | (char *) cached->data | |
976 | - (char *) db->head, cached->recsize); | |
eac10791 UD |
977 | # ifndef __ASSUME_SENDFILE |
978 | if (nwritten == -1 && errno == ENOSYS) | |
979 | goto use_write; | |
980 | # endif | |
981 | } | |
982 | else | |
983 | # ifndef __ASSUME_SENDFILE | |
984 | use_write: | |
985 | # endif | |
986 | #endif | |
987 | nwritten = writeall (fd, cached->data, cached->recsize); | |
988 | ||
989 | if (nwritten != cached->recsize | |
23700036 | 990 | && __builtin_expect (debug_level, 0) > 0) |
67479a70 UD |
991 | { |
992 | /* We have problems sending the result. */ | |
993 | char buf[256]; | |
994 | dbg_log (_("cannot write result: %s"), | |
995 | strerror_r (errno, buf, sizeof (buf))); | |
996 | } | |
997 | ||
998 | pthread_rwlock_unlock (&db->lock); | |
999 | ||
1000 | return; | |
1001 | } | |
1002 | ||
1003 | pthread_rwlock_unlock (&db->lock); | |
d67281a7 | 1004 | } |
23700036 | 1005 | else if (__builtin_expect (debug_level, 0) > 0) |
756409c4 UD |
1006 | { |
1007 | if (req->type == INVALIDATE) | |
c207f23b | 1008 | dbg_log ("\t%s (%s)", serv2str[req->type], (char *) key); |
0532e21b | 1009 | else |
a95a08b4 | 1010 | dbg_log ("\t%s", serv2str[req->type]); |
756409c4 | 1011 | } |
67479a70 UD |
1012 | |
1013 | /* Handle the request. */ | |
1014 | switch (req->type) | |
d67281a7 | 1015 | { |
67479a70 | 1016 | case GETPWBYNAME: |
a95a08b4 | 1017 | addpwbyname (db, fd, req, key, uid); |
67479a70 UD |
1018 | break; |
1019 | ||
1020 | case GETPWBYUID: | |
a95a08b4 | 1021 | addpwbyuid (db, fd, req, key, uid); |
67479a70 UD |
1022 | break; |
1023 | ||
1024 | case GETGRBYNAME: | |
a95a08b4 | 1025 | addgrbyname (db, fd, req, key, uid); |
67479a70 UD |
1026 | break; |
1027 | ||
1028 | case GETGRBYGID: | |
a95a08b4 | 1029 | addgrbygid (db, fd, req, key, uid); |
67479a70 UD |
1030 | break; |
1031 | ||
1032 | case GETHOSTBYNAME: | |
a95a08b4 | 1033 | addhstbyname (db, fd, req, key, uid); |
67479a70 UD |
1034 | break; |
1035 | ||
1036 | case GETHOSTBYNAMEv6: | |
a95a08b4 | 1037 | addhstbynamev6 (db, fd, req, key, uid); |
67479a70 UD |
1038 | break; |
1039 | ||
1040 | case GETHOSTBYADDR: | |
a95a08b4 | 1041 | addhstbyaddr (db, fd, req, key, uid); |
67479a70 UD |
1042 | break; |
1043 | ||
1044 | case GETHOSTBYADDRv6: | |
a95a08b4 | 1045 | addhstbyaddrv6 (db, fd, req, key, uid); |
67479a70 UD |
1046 | break; |
1047 | ||
d19687d6 UD |
1048 | case GETAI: |
1049 | addhstai (db, fd, req, key, uid); | |
1050 | break; | |
1051 | ||
f7e7a396 UD |
1052 | case INITGROUPS: |
1053 | addinitgroups (db, fd, req, key, uid); | |
1054 | break; | |
1055 | ||
67479a70 | 1056 | case GETSTAT: |
67479a70 | 1057 | case SHUTDOWN: |
756409c4 | 1058 | case INVALIDATE: |
70e2ebba UD |
1059 | { |
1060 | /* Get the callers credentials. */ | |
cedc8559 | 1061 | #ifdef SO_PEERCRED |
70e2ebba UD |
1062 | struct ucred caller; |
1063 | socklen_t optlen = sizeof (caller); | |
be3c40b6 | 1064 | |
70e2ebba UD |
1065 | if (getsockopt (fd, SOL_SOCKET, SO_PEERCRED, &caller, &optlen) < 0) |
1066 | { | |
1067 | char buf[256]; | |
a1c542bf | 1068 | |
70e2ebba UD |
1069 | dbg_log (_("error getting callers id: %s"), |
1070 | strerror_r (errno, buf, sizeof (buf))); | |
1071 | break; | |
1072 | } | |
a12ce44f | 1073 | |
70e2ebba | 1074 | uid = caller.uid; |
a12ce44f | 1075 | #else |
70e2ebba UD |
1076 | /* Some systems have no SO_PEERCRED implementation. They don't |
1077 | care about security so we don't as well. */ | |
1078 | uid = 0; | |
cedc8559 | 1079 | #endif |
70e2ebba | 1080 | } |
a12ce44f UD |
1081 | |
1082 | /* Accept shutdown, getstat and invalidate only from root. For | |
1083 | the stat call also allow the user specified in the config file. */ | |
1084 | if (req->type == GETSTAT) | |
1085 | { | |
1086 | if (uid == 0 || uid == stat_uid) | |
1087 | send_stats (fd, dbs); | |
1088 | } | |
1089 | else if (uid == 0) | |
1090 | { | |
1091 | if (req->type == INVALIDATE) | |
a334319f | 1092 | invalidate_cache (key); |
a12ce44f UD |
1093 | else |
1094 | termination_handler (0); | |
a1c542bf | 1095 | } |
67479a70 UD |
1096 | break; |
1097 | ||
c207f23b UD |
1098 | case GETFDPW: |
1099 | case GETFDGR: | |
1100 | case GETFDHST: | |
1101 | #ifdef SCM_RIGHTS | |
1102 | send_ro_fd (serv2db[req->type], key, fd); | |
1103 | #endif | |
1104 | break; | |
1105 | ||
67479a70 | 1106 | default: |
64acf8ed UD |
1107 | /* Ignore the command, it's nothing we know. */ |
1108 | break; | |
d67281a7 | 1109 | } |
67479a70 UD |
1110 | } |
1111 | ||
d67281a7 | 1112 | |
4401d759 UD |
1113 | /* Restart the process. */ |
1114 | static void | |
1115 | restart (void) | |
1116 | { | |
1117 | /* First determine the parameters. We do not use the parameters | |
1118 | passed to main() since in case nscd is started by running the | |
1119 | dynamic linker this will not work. Yes, this is not the usual | |
1120 | case but nscd is part of glibc and we occasionally do this. */ | |
1121 | size_t buflen = 1024; | |
1122 | char *buf = alloca (buflen); | |
1123 | size_t readlen = 0; | |
1124 | int fd = open ("/proc/self/cmdline", O_RDONLY); | |
1125 | if (fd == -1) | |
1126 | { | |
1127 | dbg_log (_("\ | |
1128 | cannot open /proc/self/cmdline: %s; disabling paranoia mode"), | |
1129 | strerror (errno)); | |
1130 | ||
1131 | paranoia = 0; | |
1132 | return; | |
1133 | } | |
1134 | ||
1135 | while (1) | |
1136 | { | |
1137 | ssize_t n = TEMP_FAILURE_RETRY (read (fd, buf + readlen, | |
1138 | buflen - readlen)); | |
1139 | if (n == -1) | |
1140 | { | |
1141 | dbg_log (_("\ | |
1142 | cannot open /proc/self/cmdline: %s; disabling paranoia mode"), | |
1143 | strerror (errno)); | |
1144 | ||
1145 | close (fd); | |
1146 | paranoia = 0; | |
1147 | return; | |
1148 | } | |
1149 | ||
1150 | readlen += n; | |
1151 | ||
1152 | if (readlen < buflen) | |
1153 | break; | |
1154 | ||
1155 | /* We might have to extend the buffer. */ | |
1156 | size_t old_buflen = buflen; | |
1157 | char *newp = extend_alloca (buf, buflen, 2 * buflen); | |
1158 | buf = memmove (newp, buf, old_buflen); | |
1159 | } | |
1160 | ||
1161 | close (fd); | |
1162 | ||
1163 | /* Parse the command line. Worst case scenario: every two | |
1164 | characters form one parameter (one character plus NUL). */ | |
1165 | char **argv = alloca ((readlen / 2 + 1) * sizeof (argv[0])); | |
1166 | int argc = 0; | |
1167 | ||
1168 | char *cp = buf; | |
1169 | while (cp < buf + readlen) | |
1170 | { | |
1171 | argv[argc++] = cp; | |
1172 | cp = (char *) rawmemchr (cp, '\0') + 1; | |
1173 | } | |
1174 | argv[argc] = NULL; | |
1175 | ||
1176 | /* Second, change back to the old user if we changed it. */ | |
1177 | if (server_user != NULL) | |
1178 | { | |
a334319f | 1179 | if (setuid (old_uid) != 0) |
4401d759 UD |
1180 | { |
1181 | dbg_log (_("\ | |
1182 | cannot change to old UID: %s; disabling paranoia mode"), | |
1183 | strerror (errno)); | |
1184 | ||
1185 | paranoia = 0; | |
1186 | return; | |
1187 | } | |
1188 | ||
a334319f | 1189 | if (setgid (old_gid) != 0) |
4401d759 UD |
1190 | { |
1191 | dbg_log (_("\ | |
1192 | cannot change to old GID: %s; disabling paranoia mode"), | |
1193 | strerror (errno)); | |
1194 | ||
1195 | setuid (server_uid); | |
1196 | paranoia = 0; | |
1197 | return; | |
1198 | } | |
1199 | } | |
1200 | ||
1201 | /* Next change back to the old working directory. */ | |
1202 | if (chdir (oldcwd) == -1) | |
1203 | { | |
1204 | dbg_log (_("\ | |
1205 | cannot change to old working directory: %s; disabling paranoia mode"), | |
1206 | strerror (errno)); | |
1207 | ||
1208 | if (server_user != NULL) | |
1209 | { | |
1210 | setuid (server_uid); | |
1211 | setgid (server_gid); | |
1212 | } | |
1213 | paranoia = 0; | |
1214 | return; | |
1215 | } | |
1216 | ||
1217 | /* Synchronize memory. */ | |
1218 | for (int cnt = 0; cnt < lastdb; ++cnt) | |
1219 | { | |
1220 | /* Make sure nobody keeps using the database. */ | |
1221 | dbs[cnt].head->timestamp = 0; | |
1222 | ||
1223 | if (dbs[cnt].persistent) | |
1224 | // XXX async OK? | |
1225 | msync (dbs[cnt].head, dbs[cnt].memsize, MS_ASYNC); | |
1226 | } | |
1227 | ||
1228 | /* The preparations are done. */ | |
1229 | execv ("/proc/self/exe", argv); | |
1230 | ||
1231 | /* If we come here, we will never be able to re-exec. */ | |
1232 | dbg_log (_("re-exec failed: %s; disabling paranoia mode"), | |
1233 | strerror (errno)); | |
1234 | ||
1235 | if (server_user != NULL) | |
1236 | { | |
1237 | setuid (server_uid); | |
1238 | setgid (server_gid); | |
1239 | } | |
233399bc UD |
1240 | if (chdir ("/") != 0) |
1241 | dbg_log (_("cannot change current working directory to \"/\": %s"), | |
1242 | strerror (errno)); | |
4401d759 UD |
1243 | paranoia = 0; |
1244 | } | |
1245 | ||
1246 | ||
1945c96f UD |
1247 | /* List of file descriptors. */ |
1248 | struct fdlist | |
1249 | { | |
1250 | int fd; | |
1251 | struct fdlist *next; | |
1252 | }; | |
1253 | /* Memory allocated for the list. */ | |
1254 | static struct fdlist *fdlist; | |
1255 | /* List of currently ready-to-read file descriptors. */ | |
1256 | static struct fdlist *readylist; | |
1257 | ||
1258 | /* Conditional variable and mutex to signal availability of entries in | |
1259 | READYLIST. The condvar is initialized dynamically since we might | |
1260 | use a different clock depending on availability. */ | |
1261 | static pthread_cond_t readylist_cond; | |
1262 | static pthread_mutex_t readylist_lock = PTHREAD_MUTEX_INITIALIZER; | |
1263 | ||
1264 | /* The clock to use with the condvar. */ | |
1265 | static clockid_t timeout_clock = CLOCK_REALTIME; | |
1266 | ||
1267 | /* Number of threads ready to handle the READYLIST. */ | |
1268 | static unsigned long int nready; | |
1269 | ||
1270 | ||
67479a70 UD |
1271 | /* This is the main loop. It is replicated in different threads but the |
1272 | `poll' call makes sure only one thread handles an incoming connection. */ | |
1273 | static void * | |
1274 | __attribute__ ((__noreturn__)) | |
1275 | nscd_run (void *p) | |
1276 | { | |
1945c96f UD |
1277 | const long int my_number = (long int) p; |
1278 | const int run_prune = my_number < lastdb && dbs[my_number].enabled; | |
1279 | struct timespec prune_ts; | |
1280 | int to = 0; | |
1281 | char buf[256]; | |
264d5b94 | 1282 | |
72ae1e97 | 1283 | if (run_prune) |
1945c96f UD |
1284 | { |
1285 | setup_thread (&dbs[my_number]); | |
081fc592 | 1286 | |
1945c96f UD |
1287 | /* We are running. */ |
1288 | dbs[my_number].head->timestamp = time (NULL); | |
d67281a7 | 1289 | |
1945c96f UD |
1290 | if (clock_gettime (timeout_clock, &prune_ts) == -1) |
1291 | /* Should never happen. */ | |
1292 | abort (); | |
0fdb4f42 | 1293 | |
1945c96f UD |
1294 | /* Compute timeout time. */ |
1295 | prune_ts.tv_sec += CACHE_PRUNE_INTERVAL; | |
1296 | } | |
1297 | ||
1298 | /* Initial locking. */ | |
1299 | pthread_mutex_lock (&readylist_lock); | |
1300 | ||
1301 | /* One more thread available. */ | |
1302 | ++nready; | |
0fdb4f42 | 1303 | |
1945c96f UD |
1304 | while (1) |
1305 | { | |
1306 | while (readylist == NULL) | |
d6db0975 | 1307 | { |
d6db0975 UD |
1308 | if (run_prune) |
1309 | { | |
1945c96f UD |
1310 | /* Wait, but not forever. */ |
1311 | to = pthread_cond_timedwait (&readylist_cond, &readylist_lock, | |
1312 | &prune_ts); | |
1313 | ||
1314 | /* If we were woken and there is no work to be done, | |
1315 | just start pruning. */ | |
1316 | if (readylist == NULL && to == ETIMEDOUT) | |
1317 | { | |
4401d759 | 1318 | --nready; |
1945c96f UD |
1319 | pthread_mutex_unlock (&readylist_lock); |
1320 | goto only_prune; | |
1321 | } | |
d6db0975 | 1322 | } |
1945c96f UD |
1323 | else |
1324 | /* No need to timeout. */ | |
1325 | pthread_cond_wait (&readylist_cond, &readylist_lock); | |
1326 | } | |
0fdb4f42 | 1327 | |
1945c96f UD |
1328 | struct fdlist *it = readylist->next; |
1329 | if (readylist->next == readylist) | |
1330 | /* Just one entry on the list. */ | |
1331 | readylist = NULL; | |
1332 | else | |
1333 | readylist->next = it->next; | |
0fdb4f42 | 1334 | |
1945c96f UD |
1335 | /* Extract the information and mark the record ready to be used |
1336 | again. */ | |
1337 | int fd = it->fd; | |
1338 | it->next = NULL; | |
0fdb4f42 | 1339 | |
1945c96f UD |
1340 | /* One more thread available. */ |
1341 | --nready; | |
67479a70 | 1342 | |
1945c96f UD |
1343 | /* We are done with the list. */ |
1344 | pthread_mutex_unlock (&readylist_lock); | |
67479a70 | 1345 | |
1945c96f UD |
1346 | /* We do not want to block on a short read or so. */ |
1347 | int fl = fcntl (fd, F_GETFL); | |
1348 | if (fl == -1 || fcntl (fd, F_SETFL, fl | O_NONBLOCK) == -1) | |
1349 | goto close_and_out; | |
0fdb4f42 UD |
1350 | |
1351 | /* Now read the request. */ | |
1945c96f | 1352 | request_header req; |
0fdb4f42 UD |
1353 | if (__builtin_expect (TEMP_FAILURE_RETRY (read (fd, &req, sizeof (req))) |
1354 | != sizeof (req), 0)) | |
d67281a7 | 1355 | { |
1945c96f UD |
1356 | /* We failed to read data. Note that this also might mean we |
1357 | failed because we would have blocked. */ | |
0fdb4f42 UD |
1358 | if (debug_level > 0) |
1359 | dbg_log (_("short read while reading request: %s"), | |
1360 | strerror_r (errno, buf, sizeof (buf))); | |
1945c96f | 1361 | goto close_and_out; |
0fdb4f42 UD |
1362 | } |
1363 | ||
3c82c131 UD |
1364 | /* Check whether this is a valid request type. */ |
1365 | if (req.type < GETPWBYNAME || req.type >= LASTREQ) | |
1366 | goto close_and_out; | |
1367 | ||
0fdb4f42 UD |
1368 | /* Some systems have no SO_PEERCRED implementation. They don't |
1369 | care about security so we don't as well. */ | |
1945c96f | 1370 | uid_t uid = -1; |
c86e6aec | 1371 | #ifdef SO_PEERCRED |
1945c96f UD |
1372 | pid_t pid = 0; |
1373 | ||
70e2ebba | 1374 | if (__builtin_expect (debug_level > 0, 0)) |
0fdb4f42 UD |
1375 | { |
1376 | struct ucred caller; | |
1377 | socklen_t optlen = sizeof (caller); | |
c86e6aec | 1378 | |
0fdb4f42 UD |
1379 | if (getsockopt (fd, SOL_SOCKET, SO_PEERCRED, &caller, &optlen) == 0) |
1380 | pid = caller.pid; | |
1381 | } | |
cedc8559 | 1382 | #endif |
a1c542bf | 1383 | |
0fdb4f42 UD |
1384 | /* It should not be possible to crash the nscd with a silly |
1385 | request (i.e., a terribly large key). We limit the size to 1kb. */ | |
1945c96f | 1386 | #define MAXKEYLEN 1024 |
0fdb4f42 | 1387 | if (__builtin_expect (req.key_len, 1) < 0 |
1945c96f | 1388 | || __builtin_expect (req.key_len, 1) > MAXKEYLEN) |
0fdb4f42 UD |
1389 | { |
1390 | if (debug_level > 0) | |
1391 | dbg_log (_("key length in request too long: %d"), req.key_len); | |
0fdb4f42 UD |
1392 | } |
1393 | else | |
1394 | { | |
1395 | /* Get the key. */ | |
1945c96f | 1396 | char keybuf[MAXKEYLEN]; |
0fdb4f42 UD |
1397 | |
1398 | if (__builtin_expect (TEMP_FAILURE_RETRY (read (fd, keybuf, | |
1399 | req.key_len)) | |
1400 | != req.key_len, 0)) | |
67479a70 | 1401 | { |
1945c96f | 1402 | /* Again, this can also mean we would have blocked. */ |
98e75a1c | 1403 | if (debug_level > 0) |
0fdb4f42 UD |
1404 | dbg_log (_("short read while reading request key: %s"), |
1405 | strerror_r (errno, buf, sizeof (buf))); | |
1945c96f | 1406 | goto close_and_out; |
67479a70 | 1407 | } |
0fdb4f42 UD |
1408 | |
1409 | if (__builtin_expect (debug_level, 0) > 0) | |
67479a70 | 1410 | { |
c86e6aec | 1411 | #ifdef SO_PEERCRED |
0fdb4f42 UD |
1412 | if (pid != 0) |
1413 | dbg_log (_("\ | |
c86e6aec | 1414 | handle_request: request received (Version = %d) from PID %ld"), |
0fdb4f42 UD |
1415 | req.version, (long int) pid); |
1416 | else | |
c86e6aec | 1417 | #endif |
0fdb4f42 | 1418 | dbg_log (_("\ |
c86e6aec | 1419 | handle_request: request received (Version = %d)"), req.version); |
0fdb4f42 | 1420 | } |
c86e6aec | 1421 | |
0fdb4f42 UD |
1422 | /* Phew, we got all the data, now process it. */ |
1423 | handle_request (fd, &req, keybuf, uid); | |
d67281a7 | 1424 | } |
264d5b94 | 1425 | |
3c82c131 UD |
1426 | close_and_out: |
1427 | /* We are done. */ | |
1428 | close (fd); | |
1429 | ||
1945c96f UD |
1430 | /* Check whether we should be pruning the cache. */ |
1431 | assert (run_prune || to == 0); | |
1432 | if (to == ETIMEDOUT) | |
1433 | { | |
1434 | only_prune: | |
1435 | /* The pthread_cond_timedwait() call timed out. It is time | |
1436 | to clean up the cache. */ | |
1437 | assert (my_number < lastdb); | |
75596b98 | 1438 | prune_cache (&dbs[my_number], time (NULL)); |
1945c96f UD |
1439 | |
1440 | if (clock_gettime (timeout_clock, &prune_ts) == -1) | |
1441 | /* Should never happen. */ | |
1442 | abort (); | |
1443 | ||
1444 | /* Compute next timeout time. */ | |
1445 | prune_ts.tv_sec += CACHE_PRUNE_INTERVAL; | |
1446 | ||
1447 | /* In case the list is emtpy we do not want to run the prune | |
1448 | code right away again. */ | |
1449 | to = 0; | |
1450 | } | |
1451 | ||
1452 | /* Re-locking. */ | |
1453 | pthread_mutex_lock (&readylist_lock); | |
1454 | ||
1455 | /* One more thread available. */ | |
1456 | ++nready; | |
1457 | } | |
1458 | } | |
1459 | ||
1460 | ||
fc03df7a UD |
1461 | static unsigned int nconns; |
1462 | ||
1945c96f | 1463 | static void |
fc03df7a | 1464 | fd_ready (int fd) |
1945c96f | 1465 | { |
fc03df7a UD |
1466 | pthread_mutex_lock (&readylist_lock); |
1467 | ||
1468 | /* Find an empty entry in FDLIST. */ | |
1469 | size_t inner; | |
1470 | for (inner = 0; inner < nconns; ++inner) | |
1471 | if (fdlist[inner].next == NULL) | |
1472 | break; | |
1473 | assert (inner < nconns); | |
1474 | ||
1475 | fdlist[inner].fd = fd; | |
1476 | ||
1477 | if (readylist == NULL) | |
1478 | readylist = fdlist[inner].next = &fdlist[inner]; | |
1945c96f | 1479 | else |
fc03df7a UD |
1480 | { |
1481 | fdlist[inner].next = readylist->next; | |
1482 | readylist = readylist->next = &fdlist[inner]; | |
1483 | } | |
1484 | ||
1485 | bool do_signal = true; | |
1486 | if (__builtin_expect (nready == 0, 0)) | |
1487 | { | |
1488 | ++client_queued; | |
1489 | do_signal = false; | |
27e82856 UD |
1490 | |
1491 | /* Try to start another thread to help out. */ | |
1492 | pthread_t th; | |
1493 | if (nthreads < max_nthreads | |
1494 | && pthread_create (&th, &attr, nscd_run, | |
1495 | (void *) (long int) nthreads) == 0) | |
1496 | { | |
1497 | /* We got another thread. */ | |
1498 | ++nthreads; | |
908c9e87 | 1499 | /* The new thread might need a kick. */ |
27e82856 UD |
1500 | do_signal = true; |
1501 | } | |
1502 | ||
fc03df7a UD |
1503 | } |
1504 | ||
1505 | pthread_mutex_unlock (&readylist_lock); | |
1506 | ||
1507 | /* Tell one of the worker threads there is work to do. */ | |
1508 | if (do_signal) | |
1509 | pthread_cond_signal (&readylist_cond); | |
1510 | } | |
1945c96f | 1511 | |
fc03df7a | 1512 | |
4401d759 UD |
1513 | /* Check whether restarting should happen. */ |
1514 | static inline int | |
1515 | restart_p (time_t now) | |
1516 | { | |
1517 | return (paranoia && readylist == NULL && nready == nthreads | |
1518 | && now >= restart_time); | |
1519 | } | |
1520 | ||
1521 | ||
1522 | /* Array for times a connection was accepted. */ | |
fc03df7a UD |
1523 | static time_t *starttime; |
1524 | ||
1525 | ||
1526 | static void | |
1527 | __attribute__ ((__noreturn__)) | |
1528 | main_loop_poll (void) | |
1529 | { | |
1945c96f UD |
1530 | struct pollfd *conns = (struct pollfd *) xmalloc (nconns |
1531 | * sizeof (conns[0])); | |
1532 | ||
1945c96f UD |
1533 | conns[0].fd = sock; |
1534 | conns[0].events = POLLRDNORM; | |
1535 | size_t nused = 1; | |
1536 | size_t firstfree = 1; | |
1537 | ||
1538 | while (1) | |
1539 | { | |
1540 | /* Wait for any event. We wait at most a couple of seconds so | |
1541 | that we can check whether we should close any of the accepted | |
1542 | connections since we have not received a request. */ | |
1543 | #define MAX_ACCEPT_TIMEOUT 30 | |
1544 | #define MIN_ACCEPT_TIMEOUT 5 | |
1545 | #define MAIN_THREAD_TIMEOUT \ | |
1546 | (MAX_ACCEPT_TIMEOUT * 1000 \ | |
1547 | - ((MAX_ACCEPT_TIMEOUT - MIN_ACCEPT_TIMEOUT) * 1000 * nused) / (2 * nconns)) | |
1548 | ||
1549 | int n = poll (conns, nused, MAIN_THREAD_TIMEOUT); | |
1550 | ||
1551 | time_t now = time (NULL); | |
1552 | ||
1553 | /* If there is a descriptor ready for reading or there is a new | |
1554 | connection, process this now. */ | |
1555 | if (n > 0) | |
67479a70 | 1556 | { |
1945c96f UD |
1557 | if (conns[0].revents != 0) |
1558 | { | |
1559 | /* We have a new incoming connection. Accept the connection. */ | |
1560 | int fd = TEMP_FAILURE_RETRY (accept (sock, NULL, NULL)); | |
1561 | ||
9388dcbb UD |
1562 | /* Use the descriptor if we have not reached the limit. */ |
1563 | if (fd >= 0) | |
1945c96f | 1564 | { |
9388dcbb UD |
1565 | if (firstfree < nconns) |
1566 | { | |
1567 | conns[firstfree].fd = fd; | |
1568 | conns[firstfree].events = POLLRDNORM; | |
1569 | starttime[firstfree] = now; | |
1570 | if (firstfree >= nused) | |
1571 | nused = firstfree + 1; | |
1572 | ||
1573 | do | |
1574 | ++firstfree; | |
1575 | while (firstfree < nused && conns[firstfree].fd != -1); | |
1576 | } | |
1577 | else | |
1578 | /* We cannot use the connection so close it. */ | |
1579 | close (fd); | |
1945c96f UD |
1580 | } |
1581 | ||
1945c96f UD |
1582 | --n; |
1583 | } | |
1584 | ||
1585 | for (size_t cnt = 1; cnt < nused && n > 0; ++cnt) | |
1586 | if (conns[cnt].revents != 0) | |
1587 | { | |
fc03df7a | 1588 | fd_ready (conns[cnt].fd); |
1945c96f UD |
1589 | |
1590 | /* Clean up the CONNS array. */ | |
1591 | conns[cnt].fd = -1; | |
1592 | if (cnt < firstfree) | |
1593 | firstfree = cnt; | |
1594 | if (cnt == nused - 1) | |
1595 | do | |
1596 | --nused; | |
1597 | while (conns[nused - 1].fd == -1); | |
1598 | ||
1599 | --n; | |
1600 | } | |
1601 | } | |
1602 | ||
1603 | /* Now find entries which have timed out. */ | |
1604 | assert (nused > 0); | |
fc03df7a UD |
1605 | |
1606 | /* We make the timeout length depend on the number of file | |
1607 | descriptors currently used. */ | |
1945c96f UD |
1608 | #define ACCEPT_TIMEOUT \ |
1609 | (MAX_ACCEPT_TIMEOUT \ | |
1610 | - ((MAX_ACCEPT_TIMEOUT - MIN_ACCEPT_TIMEOUT) * nused) / nconns) | |
fc03df7a | 1611 | time_t laststart = now - ACCEPT_TIMEOUT; |
1945c96f | 1612 | |
fc03df7a UD |
1613 | for (size_t cnt = nused - 1; cnt > 0; --cnt) |
1614 | { | |
1945c96f UD |
1615 | if (conns[cnt].fd != -1 && starttime[cnt] < laststart) |
1616 | { | |
1617 | /* Remove the entry, it timed out. */ | |
1618 | (void) close (conns[cnt].fd); | |
1619 | conns[cnt].fd = -1; | |
1620 | ||
1621 | if (cnt < firstfree) | |
1622 | firstfree = cnt; | |
1623 | if (cnt == nused - 1) | |
1624 | do | |
1625 | --nused; | |
1626 | while (conns[nused - 1].fd == -1); | |
1627 | } | |
67479a70 | 1628 | } |
4401d759 UD |
1629 | |
1630 | if (restart_p (now)) | |
1631 | restart (); | |
67308730 | 1632 | } |
d67281a7 UD |
1633 | } |
1634 | ||
67479a70 | 1635 | |
fc03df7a UD |
1636 | #ifdef HAVE_EPOLL |
1637 | static void | |
1638 | main_loop_epoll (int efd) | |
1639 | { | |
1640 | struct epoll_event ev = { 0, }; | |
1641 | int nused = 1; | |
1642 | size_t highest = 0; | |
1643 | ||
1644 | /* Add the socket. */ | |
1645 | ev.events = EPOLLRDNORM; | |
1646 | ev.data.fd = sock; | |
1647 | if (epoll_ctl (efd, EPOLL_CTL_ADD, sock, &ev) == -1) | |
1648 | /* We cannot use epoll. */ | |
1649 | return; | |
1650 | ||
1651 | while (1) | |
1652 | { | |
1653 | struct epoll_event revs[100]; | |
1654 | # define nrevs (sizeof (revs) / sizeof (revs[0])) | |
1655 | ||
1656 | int n = epoll_wait (efd, revs, nrevs, MAIN_THREAD_TIMEOUT); | |
1657 | ||
1658 | time_t now = time (NULL); | |
1659 | ||
1660 | for (int cnt = 0; cnt < n; ++cnt) | |
1661 | if (revs[cnt].data.fd == sock) | |
1662 | { | |
1663 | /* A new connection. */ | |
1664 | int fd = TEMP_FAILURE_RETRY (accept (sock, NULL, NULL)); | |
1665 | ||
1666 | if (fd >= 0) | |
1667 | { | |
1668 | /* Try to add the new descriptor. */ | |
1669 | ev.data.fd = fd; | |
1670 | if (fd >= nconns | |
1671 | || epoll_ctl (efd, EPOLL_CTL_ADD, fd, &ev) == -1) | |
1672 | /* The descriptor is too large or something went | |
1673 | wrong. Close the descriptor. */ | |
1674 | close (fd); | |
1675 | else | |
1676 | { | |
1677 | /* Remember when we accepted the connection. */ | |
1678 | starttime[fd] = now; | |
1679 | ||
1680 | if (fd > highest) | |
1681 | highest = fd; | |
1682 | ||
1683 | ++nused; | |
1684 | } | |
1685 | } | |
1686 | } | |
1687 | else | |
1688 | { | |
1689 | /* Remove the descriptor from the epoll descriptor. */ | |
908c9e87 | 1690 | (void) epoll_ctl (efd, EPOLL_CTL_DEL, revs[cnt].data.fd, NULL); |
fc03df7a | 1691 | |
2461e3dc | 1692 | /* Get a worker to handle the request. */ |
fc03df7a UD |
1693 | fd_ready (revs[cnt].data.fd); |
1694 | ||
1695 | /* Reset the time. */ | |
1696 | starttime[revs[cnt].data.fd] = 0; | |
1697 | if (revs[cnt].data.fd == highest) | |
1698 | do | |
1699 | --highest; | |
1700 | while (highest > 0 && starttime[highest] == 0); | |
1701 | ||
1702 | --nused; | |
1703 | } | |
1704 | ||
1705 | /* Now look for descriptors for accepted connections which have | |
1706 | no reply in too long of a time. */ | |
1707 | time_t laststart = now - ACCEPT_TIMEOUT; | |
1708 | for (int cnt = highest; cnt > STDERR_FILENO; --cnt) | |
1709 | if (cnt != sock && starttime[cnt] != 0 && starttime[cnt] < laststart) | |
1710 | { | |
1711 | /* We are waiting for this one for too long. Close it. */ | |
908c9e87 | 1712 | (void) epoll_ctl (efd, EPOLL_CTL_DEL, cnt, NULL); |
fc03df7a UD |
1713 | |
1714 | (void) close (cnt); | |
1715 | ||
1716 | starttime[cnt] = 0; | |
1717 | if (cnt == highest) | |
1718 | --highest; | |
1719 | } | |
1720 | else if (cnt != sock && starttime[cnt] == 0 && cnt == highest) | |
1721 | --highest; | |
4401d759 UD |
1722 | |
1723 | if (restart_p (now)) | |
1724 | restart (); | |
fc03df7a UD |
1725 | } |
1726 | } | |
1727 | #endif | |
1728 | ||
1729 | ||
67479a70 | 1730 | /* Start all the threads we want. The initial process is thread no. 1. */ |
d67281a7 | 1731 | void |
67479a70 | 1732 | start_threads (void) |
d67281a7 | 1733 | { |
1945c96f UD |
1734 | /* Initialize the conditional variable we will use. The only |
1735 | non-standard attribute we might use is the clock selection. */ | |
1736 | pthread_condattr_t condattr; | |
1737 | pthread_condattr_init (&condattr); | |
1738 | ||
3078cba2 UD |
1739 | #if defined _POSIX_CLOCK_SELECTION && _POSIX_CLOCK_SELECTION >= 0 \ |
1740 | && defined _POSIX_MONOTONIC_CLOCK && _POSIX_MONOTONIC_CLOCK >= 0 | |
1945c96f UD |
1741 | /* Determine whether the monotonous clock is available. */ |
1742 | struct timespec dummy; | |
3078cba2 | 1743 | # if _POSIX_MONOTONIC_CLOCK == 0 |
94d824f9 | 1744 | if (sysconf (_SC_MONOTONIC_CLOCK) > 0) |
3078cba2 UD |
1745 | # endif |
1746 | # if _POSIX_CLOCK_SELECTION == 0 | |
94d824f9 | 1747 | if (sysconf (_SC_CLOCK_SELECTION) > 0) |
3078cba2 | 1748 | # endif |
94d824f9 UD |
1749 | if (clock_getres (CLOCK_MONOTONIC, &dummy) == 0 |
1750 | && pthread_condattr_setclock (&condattr, CLOCK_MONOTONIC) == 0) | |
1751 | timeout_clock = CLOCK_MONOTONIC; | |
1945c96f | 1752 | #endif |
d67281a7 | 1753 | |
1945c96f UD |
1754 | pthread_cond_init (&readylist_cond, &condattr); |
1755 | pthread_condattr_destroy (&condattr); | |
1756 | ||
1757 | ||
1758 | /* Create the attribute for the threads. They are all created | |
1759 | detached. */ | |
67479a70 UD |
1760 | pthread_attr_init (&attr); |
1761 | pthread_attr_setdetachstate (&attr, PTHREAD_CREATE_DETACHED); | |
27e82856 UD |
1762 | /* Use 1MB stacks, twice as much for 64-bit architectures. */ |
1763 | pthread_attr_setstacksize (&attr, 1024 * 1024 * (sizeof (void *) / 4)); | |
d67281a7 | 1764 | |
67479a70 UD |
1765 | /* We allow less than LASTDB threads only for debugging. */ |
1766 | if (debug_level == 0) | |
1767 | nthreads = MAX (nthreads, lastdb); | |
d67281a7 | 1768 | |
27e82856 | 1769 | int nfailed = 0; |
1945c96f UD |
1770 | for (long int i = 0; i < nthreads; ++i) |
1771 | { | |
1772 | pthread_t th; | |
27e82856 UD |
1773 | if (pthread_create (&th, &attr, nscd_run, (void *) (i - nfailed)) != 0) |
1774 | ++nfailed; | |
1775 | } | |
1776 | if (nthreads - nfailed < lastdb) | |
1777 | { | |
1778 | /* We could not start enough threads. */ | |
1779 | dbg_log (_("could only start %d threads; terminating"), | |
1780 | nthreads - nfailed); | |
1781 | exit (1); | |
1945c96f | 1782 | } |
d67281a7 | 1783 | |
fc03df7a UD |
1784 | /* Determine how much room for descriptors we should initially |
1785 | allocate. This might need to change later if we cap the number | |
1786 | with MAXCONN. */ | |
1787 | const long int nfds = sysconf (_SC_OPEN_MAX); | |
1788 | #define MINCONN 32 | |
1789 | #define MAXCONN 16384 | |
1790 | if (nfds == -1 || nfds > MAXCONN) | |
1791 | nconns = MAXCONN; | |
1792 | else if (nfds < MINCONN) | |
1793 | nconns = MINCONN; | |
1794 | else | |
1795 | nconns = nfds; | |
1796 | ||
1797 | /* We need memory to pass descriptors on to the worker threads. */ | |
1798 | fdlist = (struct fdlist *) xcalloc (nconns, sizeof (fdlist[0])); | |
1799 | /* Array to keep track when connection was accepted. */ | |
1800 | starttime = (time_t *) xcalloc (nconns, sizeof (starttime[0])); | |
1801 | ||
1945c96f UD |
1802 | /* In the main thread we execute the loop which handles incoming |
1803 | connections. */ | |
fc03df7a UD |
1804 | #ifdef HAVE_EPOLL |
1805 | int efd = epoll_create (100); | |
1806 | if (efd != -1) | |
1807 | { | |
1808 | main_loop_epoll (efd); | |
1809 | close (efd); | |
1810 | } | |
1811 | #endif | |
1812 | ||
1813 | main_loop_poll (); | |
d67281a7 | 1814 | } |
057685e4 UD |
1815 | |
1816 | ||
1817 | /* Look up the uid, gid, and supplementary groups to run nscd as. When | |
1818 | this function is called, we are not listening on the nscd socket yet so | |
1819 | we can just use the ordinary lookup functions without causing a lockup */ | |
1820 | static void | |
1821 | begin_drop_privileges (void) | |
1822 | { | |
a95a08b4 | 1823 | struct passwd *pwd = getpwnam (server_user); |
057685e4 UD |
1824 | |
1825 | if (pwd == NULL) | |
1826 | { | |
1827 | dbg_log (_("Failed to run nscd as user '%s'"), server_user); | |
1828 | error (EXIT_FAILURE, 0, _("Failed to run nscd as user '%s'"), | |
1829 | server_user); | |
1830 | } | |
1831 | ||
1832 | server_uid = pwd->pw_uid; | |
1833 | server_gid = pwd->pw_gid; | |
1834 | ||
4401d759 UD |
1835 | /* Save the old UID/GID if we have to change back. */ |
1836 | if (paranoia) | |
1837 | { | |
1838 | old_uid = getuid (); | |
1839 | old_gid = getgid (); | |
1840 | } | |
1841 | ||
a95a08b4 UD |
1842 | if (getgrouplist (server_user, server_gid, NULL, &server_ngroups) == 0) |
1843 | { | |
1844 | /* This really must never happen. */ | |
1845 | dbg_log (_("Failed to run nscd as user '%s'"), server_user); | |
1846 | error (EXIT_FAILURE, errno, _("initial getgrouplist failed")); | |
1847 | } | |
057685e4 | 1848 | |
a95a08b4 | 1849 | server_groups = (gid_t *) xmalloc (server_ngroups * sizeof (gid_t)); |
057685e4 UD |
1850 | |
1851 | if (getgrouplist (server_user, server_gid, server_groups, &server_ngroups) | |
1852 | == -1) | |
1853 | { | |
1854 | dbg_log (_("Failed to run nscd as user '%s'"), server_user); | |
1855 | error (EXIT_FAILURE, errno, _("getgrouplist failed")); | |
1856 | } | |
1857 | } | |
1858 | ||
1859 | ||
1860 | /* Call setgroups(), setgid(), and setuid() to drop root privileges and | |
1861 | run nscd as the user specified in the configuration file. */ | |
1862 | static void | |
1863 | finish_drop_privileges (void) | |
1864 | { | |
1f063dca UD |
1865 | #if defined HAVE_LIBAUDIT && defined HAVE_LIBCAP |
1866 | /* We need to preserve the capabilities to connect to the audit daemon. */ | |
1867 | cap_t new_caps = preserve_capabilities (); | |
1868 | #endif | |
1869 | ||
057685e4 UD |
1870 | if (setgroups (server_ngroups, server_groups) == -1) |
1871 | { | |
1872 | dbg_log (_("Failed to run nscd as user '%s'"), server_user); | |
1873 | error (EXIT_FAILURE, errno, _("setgroups failed")); | |
1874 | } | |
1875 | ||
a334319f | 1876 | if (setgid (server_gid) == -1) |
057685e4 UD |
1877 | { |
1878 | dbg_log (_("Failed to run nscd as user '%s'"), server_user); | |
1879 | perror ("setgid"); | |
64d64de6 | 1880 | exit (4); |
057685e4 UD |
1881 | } |
1882 | ||
a334319f | 1883 | if (setuid (server_uid) == -1) |
057685e4 UD |
1884 | { |
1885 | dbg_log (_("Failed to run nscd as user '%s'"), server_user); | |
1886 | perror ("setuid"); | |
64d64de6 | 1887 | exit (4); |
057685e4 | 1888 | } |
1f063dca UD |
1889 | |
1890 | #if defined HAVE_LIBAUDIT && defined HAVE_LIBCAP | |
1891 | /* Remove the temporary capabilities. */ | |
1892 | install_real_capabilities (new_caps); | |
1893 | #endif | |
057685e4 | 1894 | } |