]>
Commit | Line | Data |
---|---|---|
0ecb606c JJ |
1 | /* Copyright (C) 1998-2002,2003,2004,2005,2006,2007 |
2 | Free Software Foundation, Inc. | |
c207f23b UD |
3 | This file is part of the GNU C Library. |
4 | Contributed by Ulrich Drepper <drepper@cygnus.com>, 1998. | |
5 | ||
6 | The GNU C Library is free software; you can redistribute it and/or | |
7 | modify it under the terms of the GNU Lesser General Public | |
8 | License as published by the Free Software Foundation; either | |
9 | version 2.1 of the License, or (at your option) any later version. | |
10 | ||
11 | The GNU C Library is distributed in the hope that it will be useful, | |
12 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
14 | Lesser General Public License for more details. | |
15 | ||
16 | You should have received a copy of the GNU Lesser General Public | |
17 | License along with the GNU C Library; if not, write to the Free | |
18 | Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA | |
19 | 02111-1307 USA. */ | |
20 | ||
21 | #include <assert.h> | |
22 | #include <errno.h> | |
23 | #include <fcntl.h> | |
24 | #include <stdbool.h> | |
0ecb606c JJ |
25 | #include <string.h> |
26 | #include <time.h> | |
c207f23b UD |
27 | #include <unistd.h> |
28 | #include <sys/mman.h> | |
29 | #include <sys/poll.h> | |
30 | #include <sys/socket.h> | |
31 | #include <sys/stat.h> | |
0ecb606c | 32 | #include <sys/time.h> |
c207f23b UD |
33 | #include <sys/uio.h> |
34 | #include <sys/un.h> | |
35 | #include <not-cancel.h> | |
36 | #include <nis/rpcsvc/nis.h> | |
37 | ||
38 | #include "nscd-client.h" | |
39 | ||
40 | ||
0ecb606c JJ |
41 | ssize_t |
42 | __readall (int fd, void *buf, size_t len) | |
43 | { | |
44 | size_t n = len; | |
45 | ssize_t ret; | |
46 | do | |
47 | { | |
48 | ret = TEMP_FAILURE_RETRY (__read (fd, buf, n)); | |
49 | if (ret <= 0) | |
50 | break; | |
51 | buf = (char *) buf + ret; | |
52 | n -= ret; | |
53 | } | |
54 | while (n > 0); | |
55 | return ret < 0 ? ret : len - n; | |
56 | } | |
57 | ||
58 | ||
59 | ssize_t | |
60 | __readvall (int fd, const struct iovec *iov, int iovcnt) | |
61 | { | |
62 | ssize_t ret = TEMP_FAILURE_RETRY (__readv (fd, iov, iovcnt)); | |
63 | if (ret <= 0) | |
64 | return ret; | |
65 | ||
66 | size_t total = 0; | |
67 | for (int i = 0; i < iovcnt; ++i) | |
68 | total += iov[i].iov_len; | |
69 | ||
70 | if (ret < total) | |
71 | { | |
72 | struct iovec iov_buf[iovcnt]; | |
73 | ssize_t r = ret; | |
74 | ||
75 | struct iovec *iovp = memcpy (iov_buf, iov, iovcnt * sizeof (*iov)); | |
76 | do | |
77 | { | |
78 | while (iovp->iov_len <= r) | |
79 | { | |
80 | r -= iovp->iov_len; | |
81 | --iovcnt; | |
82 | ++iovp; | |
83 | } | |
84 | iovp->iov_base = (char *) iovp->iov_base + r; | |
85 | iovp->iov_len -= r; | |
86 | r = TEMP_FAILURE_RETRY (__readv (fd, iovp, iovcnt)); | |
87 | if (r <= 0) | |
88 | break; | |
89 | ret += r; | |
90 | } | |
91 | while (ret < total); | |
92 | if (r < 0) | |
93 | ret = r; | |
94 | } | |
95 | return ret; | |
96 | } | |
97 | ||
98 | ||
c207f23b UD |
99 | static int |
100 | open_socket (void) | |
101 | { | |
102 | int sock = __socket (PF_UNIX, SOCK_STREAM, 0); | |
103 | if (sock < 0) | |
104 | return -1; | |
105 | ||
106 | /* Make socket non-blocking. */ | |
107 | int fl = __fcntl (sock, F_GETFL); | |
108 | if (fl != -1) | |
109 | __fcntl (sock, F_SETFL, fl | O_NONBLOCK); | |
110 | ||
111 | struct sockaddr_un sun; | |
112 | sun.sun_family = AF_UNIX; | |
113 | strcpy (sun.sun_path, _PATH_NSCDSOCKET); | |
114 | if (__connect (sock, (struct sockaddr *) &sun, sizeof (sun)) < 0 | |
115 | && errno != EINPROGRESS) | |
116 | goto out; | |
117 | ||
118 | struct pollfd fds[1]; | |
119 | fds[0].fd = sock; | |
120 | fds[0].events = POLLOUT | POLLERR | POLLHUP; | |
121 | if (__poll (fds, 1, 5 * 1000) > 0) | |
122 | /* Success. We do not check for success of the connect call here. | |
123 | If it failed, the following operations will fail. */ | |
124 | return sock; | |
125 | ||
126 | out: | |
127 | close_not_cancel_no_status (sock); | |
128 | ||
129 | return -1; | |
130 | } | |
131 | ||
132 | ||
133 | void | |
134 | __nscd_unmap (struct mapped_database *mapped) | |
135 | { | |
136 | assert (mapped->counter == 0); | |
62417d7e | 137 | __munmap ((void *) mapped->head, mapped->mapsize); |
c207f23b UD |
138 | free (mapped); |
139 | } | |
140 | ||
141 | ||
0ecb606c JJ |
142 | static int |
143 | wait_on_socket (int sock) | |
144 | { | |
145 | struct pollfd fds[1]; | |
146 | fds[0].fd = sock; | |
147 | fds[0].events = POLLIN | POLLERR | POLLHUP; | |
148 | int n = __poll (fds, 1, 5 * 1000); | |
149 | if (n == -1 && __builtin_expect (errno == EINTR, 0)) | |
150 | { | |
151 | /* Handle the case where the poll() call is interrupted by a | |
152 | signal. We cannot just use TEMP_FAILURE_RETRY since it might | |
153 | lead to infinite loops. */ | |
154 | struct timeval now; | |
155 | (void) __gettimeofday (&now, NULL); | |
156 | long int end = (now.tv_sec + 5) * 1000 + (now.tv_usec + 500) / 1000; | |
157 | while (1) | |
158 | { | |
159 | long int timeout = end - (now.tv_sec * 1000 | |
160 | + (now.tv_usec + 500) / 1000); | |
161 | n = __poll (fds, 1, timeout); | |
162 | if (n != -1 || errno != EINTR) | |
163 | break; | |
164 | (void) __gettimeofday (&now, NULL); | |
165 | } | |
166 | } | |
167 | ||
168 | return n; | |
169 | } | |
170 | ||
171 | ||
c207f23b UD |
172 | /* Try to get a file descriptor for the shared meory segment |
173 | containing the database. */ | |
174 | static struct mapped_database * | |
175 | get_mapping (request_type type, const char *key, | |
176 | struct mapped_database **mappedp) | |
177 | { | |
178 | struct mapped_database *result = NO_MAPPING; | |
179 | #ifdef SCM_RIGHTS | |
180 | const size_t keylen = strlen (key) + 1; | |
c207f23b UD |
181 | int saved_errno = errno; |
182 | ||
183 | int mapfd = -1; | |
184 | ||
185 | /* Send the request. */ | |
0ecb606c JJ |
186 | struct |
187 | { | |
188 | request_header req; | |
189 | char key[keylen]; | |
190 | } reqdata; | |
191 | size_t real_sizeof_reqdata = sizeof (request_header) + keylen; | |
c207f23b UD |
192 | |
193 | int sock = open_socket (); | |
194 | if (sock < 0) | |
195 | goto out; | |
196 | ||
0ecb606c JJ |
197 | reqdata.req.version = NSCD_VERSION; |
198 | reqdata.req.type = type; | |
199 | reqdata.req.key_len = keylen; | |
200 | memcpy (reqdata.key, key, keylen); | |
201 | ||
202 | # ifndef MSG_NOSIGNAL | |
203 | # define MSG_NOSIGNAL 0 | |
204 | # endif | |
205 | if (__builtin_expect (TEMP_FAILURE_RETRY (__send (sock, &reqdata, | |
206 | real_sizeof_reqdata, | |
207 | MSG_NOSIGNAL)) | |
208 | != real_sizeof_reqdata, 0)) | |
c207f23b UD |
209 | /* We cannot even write the request. */ |
210 | goto out_close2; | |
211 | ||
212 | /* Room for the data sent along with the file descriptor. We expect | |
213 | the key name back. */ | |
0ecb606c JJ |
214 | # define resdata reqdata.key |
215 | struct iovec iov[1]; | |
c207f23b UD |
216 | iov[0].iov_base = resdata; |
217 | iov[0].iov_len = keylen; | |
218 | ||
0ecb606c JJ |
219 | union |
220 | { | |
221 | struct cmsghdr hdr; | |
222 | char bytes[CMSG_SPACE (sizeof (int))]; | |
223 | } buf; | |
c207f23b | 224 | struct msghdr msg = { .msg_iov = iov, .msg_iovlen = 1, |
0ecb606c JJ |
225 | .msg_control = buf.bytes, |
226 | .msg_controllen = sizeof (buf) }; | |
c207f23b UD |
227 | struct cmsghdr *cmsg = CMSG_FIRSTHDR (&msg); |
228 | ||
229 | cmsg->cmsg_level = SOL_SOCKET; | |
230 | cmsg->cmsg_type = SCM_RIGHTS; | |
231 | cmsg->cmsg_len = CMSG_LEN (sizeof (int)); | |
232 | ||
0ecb606c JJ |
233 | /* This access is well-aligned since BUF is correctly aligned for an |
234 | int and CMSG_DATA preserves this alignment. */ | |
c207f23b UD |
235 | *(int *) CMSG_DATA (cmsg) = -1; |
236 | ||
237 | msg.msg_controllen = cmsg->cmsg_len; | |
238 | ||
0ecb606c | 239 | if (wait_on_socket (sock) <= 0) |
c207f23b UD |
240 | goto out_close2; |
241 | ||
0ecb606c JJ |
242 | if (__builtin_expect (TEMP_FAILURE_RETRY (__recvmsg (sock, &msg, 0)) |
243 | != keylen, 0)) | |
c207f23b UD |
244 | goto out_close2; |
245 | ||
0ecb606c JJ |
246 | if (__builtin_expect (CMSG_FIRSTHDR (&msg) == NULL |
247 | || (CMSG_FIRSTHDR (&msg)->cmsg_len | |
248 | != CMSG_LEN (sizeof (int))), 0)) | |
249 | goto out_close2; | |
aadd7d9d | 250 | |
0ecb606c | 251 | mapfd = *(int *) CMSG_DATA (cmsg); |
407c4b9a | 252 | |
c207f23b | 253 | struct stat64 st; |
0ecb606c JJ |
254 | if (__builtin_expect (strcmp (resdata, key) != 0, 0) |
255 | || __builtin_expect (fstat64 (mapfd, &st) != 0, 0) | |
256 | || __builtin_expect (st.st_size < sizeof (struct database_pers_head), 0)) | |
c207f23b UD |
257 | goto out_close; |
258 | ||
259 | struct database_pers_head head; | |
0ecb606c JJ |
260 | if (__builtin_expect (TEMP_FAILURE_RETRY (__pread (mapfd, &head, |
261 | sizeof (head), 0)) | |
262 | != sizeof (head), 0)) | |
c207f23b UD |
263 | goto out_close; |
264 | ||
0ecb606c JJ |
265 | if (__builtin_expect (head.version != DB_VERSION, 0) |
266 | || __builtin_expect (head.header_size != sizeof (head), 0) | |
c207f23b UD |
267 | /* This really should not happen but who knows, maybe the update |
268 | thread got stuck. */ | |
0ecb606c JJ |
269 | || __builtin_expect (! head.nscd_certainly_running |
270 | && head.timestamp + MAPPING_TIMEOUT < time (NULL), | |
271 | 0)) | |
c207f23b UD |
272 | goto out_close; |
273 | ||
274 | size_t size = (sizeof (head) + roundup (head.module * sizeof (ref_t), ALIGN) | |
275 | + head.data_size); | |
276 | ||
0ecb606c | 277 | if (__builtin_expect (st.st_size < size, 0)) |
c207f23b UD |
278 | goto out_close; |
279 | ||
280 | /* The file is large enough, map it now. */ | |
281 | void *mapping = __mmap (NULL, size, PROT_READ, MAP_SHARED, mapfd, 0); | |
0ecb606c | 282 | if (__builtin_expect (mapping != MAP_FAILED, 1)) |
c207f23b UD |
283 | { |
284 | /* Allocate a record for the mapping. */ | |
0ecb606c | 285 | struct mapped_database *newp = malloc (sizeof (*newp)); |
c207f23b UD |
286 | if (newp == NULL) |
287 | { | |
288 | /* Ugh, after all we went through the memory allocation failed. */ | |
0ecb606c | 289 | __munmap (mapping, size); |
c207f23b UD |
290 | goto out_close; |
291 | } | |
292 | ||
293 | newp->head = mapping; | |
294 | newp->data = ((char *) mapping + head.header_size | |
295 | + roundup (head.module * sizeof (ref_t), ALIGN)); | |
296 | newp->mapsize = size; | |
0ecb606c | 297 | newp->datasize = head.data_size; |
c207f23b UD |
298 | /* Set counter to 1 to show it is usable. */ |
299 | newp->counter = 1; | |
300 | ||
301 | result = newp; | |
302 | } | |
303 | ||
304 | out_close: | |
305 | __close (mapfd); | |
306 | out_close2: | |
307 | __close (sock); | |
308 | out: | |
309 | __set_errno (saved_errno); | |
310 | #endif /* SCM_RIGHTS */ | |
311 | ||
312 | struct mapped_database *oldval = *mappedp; | |
313 | *mappedp = result; | |
314 | ||
315 | if (oldval != NULL && atomic_decrement_val (&oldval->counter) == 0) | |
316 | __nscd_unmap (oldval); | |
317 | ||
318 | return result; | |
319 | } | |
320 | ||
321 | ||
322 | struct mapped_database * | |
323 | __nscd_get_map_ref (request_type type, const char *name, | |
0ecb606c | 324 | volatile struct locked_map_ptr *mapptr, int *gc_cyclep) |
c207f23b UD |
325 | { |
326 | struct mapped_database *cur = mapptr->mapped; | |
327 | if (cur == NO_MAPPING) | |
328 | return cur; | |
329 | ||
330 | int cnt = 0; | |
0ecb606c JJ |
331 | while (__builtin_expect (atomic_compare_and_exchange_val_acq (&mapptr->lock, |
332 | 1, 0) != 0, 0)) | |
c207f23b UD |
333 | { |
334 | // XXX Best number of rounds? | |
0ecb606c | 335 | if (__builtin_expect (++cnt > 5, 0)) |
c207f23b UD |
336 | return NO_MAPPING; |
337 | ||
338 | atomic_delay (); | |
339 | } | |
340 | ||
341 | cur = mapptr->mapped; | |
342 | ||
343 | if (__builtin_expect (cur != NO_MAPPING, 1)) | |
344 | { | |
345 | /* If not mapped or timestamp not updated, request new map. */ | |
346 | if (cur == NULL | |
081fc592 | 347 | || (cur->head->nscd_certainly_running == 0 |
0ecb606c JJ |
348 | && cur->head->timestamp + MAPPING_TIMEOUT < time (NULL)) |
349 | || cur->head->data_size > cur->datasize) | |
350 | cur = get_mapping (type, name, | |
351 | (struct mapped_database **) &mapptr->mapped); | |
c207f23b UD |
352 | |
353 | if (__builtin_expect (cur != NO_MAPPING, 1)) | |
354 | { | |
355 | if (__builtin_expect (((*gc_cyclep = cur->head->gc_cycle) & 1) != 0, | |
356 | 0)) | |
357 | cur = NO_MAPPING; | |
358 | else | |
359 | atomic_increment (&cur->counter); | |
360 | } | |
361 | } | |
362 | ||
363 | mapptr->lock = 0; | |
364 | ||
365 | return cur; | |
366 | } | |
367 | ||
368 | ||
0ecb606c JJ |
369 | /* Don't return const struct datahead *, as eventhough the record |
370 | is normally constant, it can change arbitrarily during nscd | |
371 | garbage collection. */ | |
372 | struct datahead * | |
c207f23b UD |
373 | __nscd_cache_search (request_type type, const char *key, size_t keylen, |
374 | const struct mapped_database *mapped) | |
375 | { | |
376 | unsigned long int hash = __nis_hash (key, keylen) % mapped->head->module; | |
0ecb606c | 377 | size_t datasize = mapped->datasize; |
c207f23b UD |
378 | |
379 | ref_t work = mapped->head->array[hash]; | |
0ecb606c | 380 | while (work != ENDREF && work + sizeof (struct hashentry) <= datasize) |
c207f23b UD |
381 | { |
382 | struct hashentry *here = (struct hashentry *) (mapped->data + work); | |
383 | ||
0ecb606c JJ |
384 | #ifndef _STRING_ARCH_unaligned |
385 | /* Although during garbage collection when moving struct hashentry | |
386 | records around we first copy from old to new location and then | |
387 | adjust pointer from previous hashentry to it, there is no barrier | |
388 | between those memory writes. It is very unlikely to hit it, | |
389 | so check alignment only if a misaligned load can crash the | |
390 | application. */ | |
391 | if ((uintptr_t) here & (__alignof__ (*here) - 1)) | |
392 | return NULL; | |
393 | #endif | |
394 | ||
395 | if (type == here->type | |
396 | && keylen == here->len | |
397 | && here->key + keylen <= datasize | |
398 | && memcmp (key, mapped->data + here->key, keylen) == 0 | |
399 | && here->packet + sizeof (struct datahead) <= datasize) | |
c207f23b UD |
400 | { |
401 | /* We found the entry. Increment the appropriate counter. */ | |
0ecb606c | 402 | struct datahead *dh |
c207f23b UD |
403 | = (struct datahead *) (mapped->data + here->packet); |
404 | ||
0ecb606c JJ |
405 | #ifndef _STRING_ARCH_unaligned |
406 | if ((uintptr_t) dh & (__alignof__ (*dh) - 1)) | |
407 | return NULL; | |
408 | #endif | |
409 | ||
c207f23b UD |
410 | /* See whether we must ignore the entry or whether something |
411 | is wrong because garbage collection is in progress. */ | |
0ecb606c | 412 | if (dh->usable && here->packet + dh->allocsize <= datasize) |
c207f23b UD |
413 | return dh; |
414 | } | |
415 | ||
416 | work = here->next; | |
417 | } | |
418 | ||
419 | return NULL; | |
420 | } | |
421 | ||
422 | ||
423 | /* Create a socket connected to a name. */ | |
424 | int | |
425 | __nscd_open_socket (const char *key, size_t keylen, request_type type, | |
426 | void *response, size_t responselen) | |
427 | { | |
428 | int saved_errno = errno; | |
429 | ||
430 | int sock = open_socket (); | |
431 | if (sock >= 0) | |
432 | { | |
433 | request_header req; | |
434 | req.version = NSCD_VERSION; | |
435 | req.type = type; | |
436 | req.key_len = keylen; | |
437 | ||
438 | struct iovec vec[2]; | |
439 | vec[0].iov_base = &req; | |
440 | vec[0].iov_len = sizeof (request_header); | |
441 | vec[1].iov_base = (void *) key; | |
442 | vec[1].iov_len = keylen; | |
443 | ||
444 | ssize_t nbytes = TEMP_FAILURE_RETRY (__writev (sock, vec, 2)); | |
0ecb606c | 445 | if (nbytes == (ssize_t) (sizeof (request_header) + keylen) |
c207f23b | 446 | /* Wait for data. */ |
0ecb606c JJ |
447 | && wait_on_socket (sock) > 0) |
448 | { | |
449 | nbytes = TEMP_FAILURE_RETRY (__read (sock, response, responselen)); | |
450 | if (nbytes == (ssize_t) responselen) | |
451 | return sock; | |
c207f23b UD |
452 | } |
453 | ||
454 | close_not_cancel_no_status (sock); | |
455 | } | |
456 | ||
457 | __set_errno (saved_errno); | |
458 | ||
459 | return -1; | |
460 | } |