1 | /*
|
---|
2 | Unix SMB/CIFS implementation.
|
---|
3 |
|
---|
4 | trivial database library
|
---|
5 |
|
---|
6 | Copyright (C) Andrew Tridgell 1999-2005
|
---|
7 | Copyright (C) Paul `Rusty' Russell 2000
|
---|
8 | Copyright (C) Jeremy Allison 2000-2003
|
---|
9 |
|
---|
10 | ** NOTE! The following LGPL license applies to the tdb
|
---|
11 | ** library. This does NOT imply that all of Samba is released
|
---|
12 | ** under the LGPL
|
---|
13 |
|
---|
14 | This library is free software; you can redistribute it and/or
|
---|
15 | modify it under the terms of the GNU Lesser General Public
|
---|
16 | License as published by the Free Software Foundation; either
|
---|
17 | version 3 of the License, or (at your option) any later version.
|
---|
18 |
|
---|
19 | This library is distributed in the hope that it will be useful,
|
---|
20 | but WITHOUT ANY WARRANTY; without even the implied warranty of
|
---|
21 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
---|
22 | Lesser General Public License for more details.
|
---|
23 |
|
---|
24 | You should have received a copy of the GNU Lesser General Public
|
---|
25 | License along with this library; if not, see <http://www.gnu.org/licenses/>.
|
---|
26 | */
|
---|
27 |
|
---|
28 | #include "tdb_private.h"
|
---|
29 |
|
---|
30 | /* all contexts, to ensure no double-opens (fcntl locks don't nest!) */
|
---|
31 | static struct tdb_context *tdbs = NULL;
|
---|
32 |
|
---|
33 | /* We use two hashes to double-check they're using the right hash function. */
|
---|
34 | void tdb_header_hash(struct tdb_context *tdb,
|
---|
35 | uint32_t *magic1_hash, uint32_t *magic2_hash)
|
---|
36 | {
|
---|
37 | TDB_DATA hash_key;
|
---|
38 | uint32_t tdb_magic = TDB_MAGIC;
|
---|
39 |
|
---|
40 | hash_key.dptr = discard_const_p(unsigned char, TDB_MAGIC_FOOD);
|
---|
41 | hash_key.dsize = sizeof(TDB_MAGIC_FOOD);
|
---|
42 | *magic1_hash = tdb->hash_fn(&hash_key);
|
---|
43 |
|
---|
44 | hash_key.dptr = (unsigned char *)CONVERT(tdb_magic);
|
---|
45 | hash_key.dsize = sizeof(tdb_magic);
|
---|
46 | *magic2_hash = tdb->hash_fn(&hash_key);
|
---|
47 |
|
---|
48 | /* Make sure at least one hash is non-zero! */
|
---|
49 | if (*magic1_hash == 0 && *magic2_hash == 0)
|
---|
50 | *magic1_hash = 1;
|
---|
51 | }
|
---|
52 |
|
---|
53 | /* initialise a new database with a specified hash size */
|
---|
54 | static int tdb_new_database(struct tdb_context *tdb, int hash_size)
|
---|
55 | {
|
---|
56 | struct tdb_header *newdb;
|
---|
57 | size_t size;
|
---|
58 | int ret = -1;
|
---|
59 |
|
---|
60 | /* We make it up in memory, then write it out if not internal */
|
---|
61 | size = sizeof(struct tdb_header) + (hash_size+1)*sizeof(tdb_off_t);
|
---|
62 | if (!(newdb = (struct tdb_header *)calloc(size, 1))) {
|
---|
63 | tdb->ecode = TDB_ERR_OOM;
|
---|
64 | return -1;
|
---|
65 | }
|
---|
66 |
|
---|
67 | /* Fill in the header */
|
---|
68 | newdb->version = TDB_VERSION;
|
---|
69 | newdb->hash_size = hash_size;
|
---|
70 |
|
---|
71 | tdb_header_hash(tdb, &newdb->magic1_hash, &newdb->magic2_hash);
|
---|
72 |
|
---|
73 | /* Make sure older tdbs (which don't check the magic hash fields)
|
---|
74 | * will refuse to open this TDB. */
|
---|
75 | if (tdb->flags & TDB_INCOMPATIBLE_HASH)
|
---|
76 | newdb->rwlocks = TDB_HASH_RWLOCK_MAGIC;
|
---|
77 |
|
---|
78 | if (tdb->flags & TDB_INTERNAL) {
|
---|
79 | tdb->map_size = size;
|
---|
80 | tdb->map_ptr = (char *)newdb;
|
---|
81 | memcpy(&tdb->header, newdb, sizeof(tdb->header));
|
---|
82 | /* Convert the `ondisk' version if asked. */
|
---|
83 | CONVERT(*newdb);
|
---|
84 | return 0;
|
---|
85 | }
|
---|
86 | if (lseek(tdb->fd, 0, SEEK_SET) == -1)
|
---|
87 | goto fail;
|
---|
88 |
|
---|
89 | if (ftruncate(tdb->fd, 0) == -1)
|
---|
90 | goto fail;
|
---|
91 |
|
---|
92 | /* This creates an endian-converted header, as if read from disk */
|
---|
93 | CONVERT(*newdb);
|
---|
94 | memcpy(&tdb->header, newdb, sizeof(tdb->header));
|
---|
95 | /* Don't endian-convert the magic food! */
|
---|
96 | memcpy(newdb->magic_food, TDB_MAGIC_FOOD, strlen(TDB_MAGIC_FOOD)+1);
|
---|
97 | /* we still have "ret == -1" here */
|
---|
98 | if (tdb_write_all(tdb->fd, newdb, size))
|
---|
99 | ret = 0;
|
---|
100 |
|
---|
101 | fail:
|
---|
102 | SAFE_FREE(newdb);
|
---|
103 | return ret;
|
---|
104 | }
|
---|
105 |
|
---|
106 |
|
---|
107 |
|
---|
108 | static int tdb_already_open(dev_t device,
|
---|
109 | ino_t ino)
|
---|
110 | {
|
---|
111 | struct tdb_context *i;
|
---|
112 |
|
---|
113 | for (i = tdbs; i; i = i->next) {
|
---|
114 | if (i->device == device && i->inode == ino) {
|
---|
115 | return 1;
|
---|
116 | }
|
---|
117 | }
|
---|
118 |
|
---|
119 | return 0;
|
---|
120 | }
|
---|
121 |
|
---|
122 | /* open the database, creating it if necessary
|
---|
123 |
|
---|
124 | The open_flags and mode are passed straight to the open call on the
|
---|
125 | database file. A flags value of O_WRONLY is invalid. The hash size
|
---|
126 | is advisory, use zero for a default value.
|
---|
127 |
|
---|
128 | Return is NULL on error, in which case errno is also set. Don't
|
---|
129 | try to call tdb_error or tdb_errname, just do strerror(errno).
|
---|
130 |
|
---|
131 | @param name may be NULL for internal databases. */
|
---|
132 | _PUBLIC_ struct tdb_context *tdb_open(const char *name, int hash_size, int tdb_flags,
|
---|
133 | int open_flags, mode_t mode)
|
---|
134 | {
|
---|
135 | return tdb_open_ex(name, hash_size, tdb_flags, open_flags, mode, NULL, NULL);
|
---|
136 | }
|
---|
137 |
|
---|
138 | /* a default logging function */
|
---|
139 | static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...) PRINTF_ATTRIBUTE(3, 4);
|
---|
140 | static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...)
|
---|
141 | {
|
---|
142 | }
|
---|
143 |
|
---|
144 | static bool check_header_hash(struct tdb_context *tdb,
|
---|
145 | bool default_hash, uint32_t *m1, uint32_t *m2)
|
---|
146 | {
|
---|
147 | tdb_header_hash(tdb, m1, m2);
|
---|
148 | if (tdb->header.magic1_hash == *m1 &&
|
---|
149 | tdb->header.magic2_hash == *m2) {
|
---|
150 | return true;
|
---|
151 | }
|
---|
152 |
|
---|
153 | /* If they explicitly set a hash, always respect it. */
|
---|
154 | if (!default_hash)
|
---|
155 | return false;
|
---|
156 |
|
---|
157 | /* Otherwise, try the other inbuilt hash. */
|
---|
158 | if (tdb->hash_fn == tdb_old_hash)
|
---|
159 | tdb->hash_fn = tdb_jenkins_hash;
|
---|
160 | else
|
---|
161 | tdb->hash_fn = tdb_old_hash;
|
---|
162 | return check_header_hash(tdb, false, m1, m2);
|
---|
163 | }
|
---|
164 |
|
---|
165 | _PUBLIC_ struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
|
---|
166 | int open_flags, mode_t mode,
|
---|
167 | const struct tdb_logging_context *log_ctx,
|
---|
168 | tdb_hash_func hash_fn)
|
---|
169 | {
|
---|
170 | struct tdb_context *tdb;
|
---|
171 | struct stat st;
|
---|
172 | int rev = 0, locked = 0;
|
---|
173 | unsigned char *vp;
|
---|
174 | uint32_t vertest;
|
---|
175 | unsigned v;
|
---|
176 | const char *hash_alg;
|
---|
177 | uint32_t magic1, magic2;
|
---|
178 |
|
---|
179 | if (!(tdb = (struct tdb_context *)calloc(1, sizeof *tdb))) {
|
---|
180 | /* Can't log this */
|
---|
181 | errno = ENOMEM;
|
---|
182 | goto fail;
|
---|
183 | }
|
---|
184 | tdb_io_init(tdb);
|
---|
185 | tdb->fd = -1;
|
---|
186 | #ifdef TDB_TRACE
|
---|
187 | tdb->tracefd = -1;
|
---|
188 | #endif
|
---|
189 | tdb->name = NULL;
|
---|
190 | tdb->map_ptr = NULL;
|
---|
191 | tdb->flags = tdb_flags;
|
---|
192 | #ifdef __OS2__
|
---|
193 | open_flags |= O_BINARY;
|
---|
194 | #endif
|
---|
195 | tdb->open_flags = open_flags;
|
---|
196 | if (log_ctx) {
|
---|
197 | tdb->log = *log_ctx;
|
---|
198 | } else {
|
---|
199 | tdb->log.log_fn = null_log_fn;
|
---|
200 | tdb->log.log_private = NULL;
|
---|
201 | }
|
---|
202 |
|
---|
203 | if (name == NULL && (tdb_flags & TDB_INTERNAL)) {
|
---|
204 | name = "__TDB_INTERNAL__";
|
---|
205 | }
|
---|
206 |
|
---|
207 | if (name == NULL) {
|
---|
208 | tdb->name = discard_const_p(char, "__NULL__");
|
---|
209 | TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: called with name == NULL\n"));
|
---|
210 | tdb->name = NULL;
|
---|
211 | errno = EINVAL;
|
---|
212 | goto fail;
|
---|
213 | }
|
---|
214 |
|
---|
215 | /* now make a copy of the name, as the caller memory might went away */
|
---|
216 | if (!(tdb->name = (char *)strdup(name))) {
|
---|
217 | /*
|
---|
218 | * set the name as the given string, so that tdb_name() will
|
---|
219 | * work in case of an error.
|
---|
220 | */
|
---|
221 | tdb->name = discard_const_p(char, name);
|
---|
222 | TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: can't strdup(%s)\n",
|
---|
223 | name));
|
---|
224 | tdb->name = NULL;
|
---|
225 | errno = ENOMEM;
|
---|
226 | goto fail;
|
---|
227 | }
|
---|
228 |
|
---|
229 | if (hash_fn) {
|
---|
230 | tdb->hash_fn = hash_fn;
|
---|
231 | hash_alg = "the user defined";
|
---|
232 | } else {
|
---|
233 | /* This controls what we use when creating a tdb. */
|
---|
234 | if (tdb->flags & TDB_INCOMPATIBLE_HASH) {
|
---|
235 | tdb->hash_fn = tdb_jenkins_hash;
|
---|
236 | } else {
|
---|
237 | tdb->hash_fn = tdb_old_hash;
|
---|
238 | }
|
---|
239 | hash_alg = "either default";
|
---|
240 | }
|
---|
241 |
|
---|
242 | /* cache the page size */
|
---|
243 | tdb->page_size = getpagesize();
|
---|
244 | if (tdb->page_size <= 0) {
|
---|
245 | tdb->page_size = 0x2000;
|
---|
246 | }
|
---|
247 |
|
---|
248 | tdb->max_dead_records = (tdb_flags & TDB_VOLATILE) ? 5 : 0;
|
---|
249 |
|
---|
250 | if ((open_flags & O_ACCMODE) == O_WRONLY) {
|
---|
251 | TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: can't open tdb %s write-only\n",
|
---|
252 | name));
|
---|
253 | errno = EINVAL;
|
---|
254 | goto fail;
|
---|
255 | }
|
---|
256 |
|
---|
257 | if (hash_size == 0)
|
---|
258 | hash_size = DEFAULT_HASH_SIZE;
|
---|
259 | if ((open_flags & O_ACCMODE) == O_RDONLY) {
|
---|
260 | tdb->read_only = 1;
|
---|
261 | /* read only databases don't do locking or clear if first */
|
---|
262 | tdb->flags |= TDB_NOLOCK;
|
---|
263 | tdb->flags &= ~TDB_CLEAR_IF_FIRST;
|
---|
264 | }
|
---|
265 |
|
---|
266 | if ((tdb->flags & TDB_ALLOW_NESTING) &&
|
---|
267 | (tdb->flags & TDB_DISALLOW_NESTING)) {
|
---|
268 | tdb->ecode = TDB_ERR_NESTING;
|
---|
269 | TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
|
---|
270 | "allow_nesting and disallow_nesting are not allowed together!"));
|
---|
271 | errno = EINVAL;
|
---|
272 | goto fail;
|
---|
273 | }
|
---|
274 |
|
---|
275 | if (getenv("TDB_NO_FSYNC")) {
|
---|
276 | tdb->flags |= TDB_NOSYNC;
|
---|
277 | }
|
---|
278 |
|
---|
279 | /*
|
---|
280 | * TDB_ALLOW_NESTING is the default behavior.
|
---|
281 | * Note: this may change in future versions!
|
---|
282 | */
|
---|
283 | if (!(tdb->flags & TDB_DISALLOW_NESTING)) {
|
---|
284 | tdb->flags |= TDB_ALLOW_NESTING;
|
---|
285 | }
|
---|
286 |
|
---|
287 | /* internal databases don't mmap or lock, and start off cleared */
|
---|
288 | if (tdb->flags & TDB_INTERNAL) {
|
---|
289 | tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
|
---|
290 | tdb->flags &= ~TDB_CLEAR_IF_FIRST;
|
---|
291 | if (tdb_new_database(tdb, hash_size) != 0) {
|
---|
292 | TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: tdb_new_database failed!"));
|
---|
293 | goto fail;
|
---|
294 | }
|
---|
295 | goto internal;
|
---|
296 | }
|
---|
297 |
|
---|
298 | if ((tdb->fd = open(name, open_flags, mode)) == -1) {
|
---|
299 | TDB_LOG((tdb, TDB_DEBUG_WARNING, "tdb_open_ex: could not open file %s: %s\n",
|
---|
300 | name, strerror(errno)));
|
---|
301 | goto fail; /* errno set by open(2) */
|
---|
302 | }
|
---|
303 |
|
---|
304 | /* on exec, don't inherit the fd */
|
---|
305 | v = fcntl(tdb->fd, F_GETFD, 0);
|
---|
306 | fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
|
---|
307 |
|
---|
308 | /* ensure there is only one process initialising at once */
|
---|
309 | if (tdb_nest_lock(tdb, OPEN_LOCK, F_WRLCK, TDB_LOCK_WAIT) == -1) {
|
---|
310 | TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to get open lock on %s: %s\n",
|
---|
311 | name, strerror(errno)));
|
---|
312 | goto fail; /* errno set by tdb_brlock */
|
---|
313 | }
|
---|
314 |
|
---|
315 | /* we need to zero database if we are the only one with it open */
|
---|
316 | if ((tdb_flags & TDB_CLEAR_IF_FIRST) &&
|
---|
317 | (!tdb->read_only) &&
|
---|
318 | (locked = (tdb_nest_lock(tdb, ACTIVE_LOCK, F_WRLCK, TDB_LOCK_NOWAIT|TDB_LOCK_PROBE) == 0))) {
|
---|
319 | int ret;
|
---|
320 | ret = tdb_brlock(tdb, F_WRLCK, FREELIST_TOP, 0,
|
---|
321 | TDB_LOCK_WAIT);
|
---|
322 | if (ret == -1) {
|
---|
323 | TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
|
---|
324 | "tdb_brlock failed for %s: %s\n",
|
---|
325 | name, strerror(errno)));
|
---|
326 | goto fail;
|
---|
327 | }
|
---|
328 | ret = tdb_new_database(tdb, hash_size);
|
---|
329 | if (ret == -1) {
|
---|
330 | TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
|
---|
331 | "tdb_new_database failed for %s: %s\n",
|
---|
332 | name, strerror(errno)));
|
---|
333 | tdb_unlockall(tdb);
|
---|
334 | goto fail;
|
---|
335 | }
|
---|
336 | ret = tdb_brunlock(tdb, F_WRLCK, FREELIST_TOP, 0);
|
---|
337 | if (ret == -1) {
|
---|
338 | TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
|
---|
339 | "tdb_unlockall failed for %s: %s\n",
|
---|
340 | name, strerror(errno)));
|
---|
341 | goto fail;
|
---|
342 | }
|
---|
343 | ret = lseek(tdb->fd, 0, SEEK_SET);
|
---|
344 | if (ret == -1) {
|
---|
345 | TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
|
---|
346 | "lseek failed for %s: %s\n",
|
---|
347 | name, strerror(errno)));
|
---|
348 | goto fail;
|
---|
349 | }
|
---|
350 | }
|
---|
351 |
|
---|
352 | errno = 0;
|
---|
353 | if (read(tdb->fd, &tdb->header, sizeof(tdb->header)) != sizeof(tdb->header)
|
---|
354 | || strcmp(tdb->header.magic_food, TDB_MAGIC_FOOD) != 0) {
|
---|
355 | if (!(open_flags & O_CREAT) || tdb_new_database(tdb, hash_size) == -1) {
|
---|
356 | if (errno == 0) {
|
---|
357 | errno = EIO; /* ie bad format or something */
|
---|
358 | }
|
---|
359 | goto fail;
|
---|
360 | }
|
---|
361 | rev = (tdb->flags & TDB_CONVERT);
|
---|
362 | } else if (tdb->header.version != TDB_VERSION
|
---|
363 | && !(rev = (tdb->header.version==TDB_BYTEREV(TDB_VERSION)))) {
|
---|
364 | /* wrong version */
|
---|
365 | errno = EIO;
|
---|
366 | goto fail;
|
---|
367 | }
|
---|
368 | vp = (unsigned char *)&tdb->header.version;
|
---|
369 | vertest = (((uint32_t)vp[0]) << 24) | (((uint32_t)vp[1]) << 16) |
|
---|
370 | (((uint32_t)vp[2]) << 8) | (uint32_t)vp[3];
|
---|
371 | tdb->flags |= (vertest==TDB_VERSION) ? TDB_BIGENDIAN : 0;
|
---|
372 | if (!rev)
|
---|
373 | tdb->flags &= ~TDB_CONVERT;
|
---|
374 | else {
|
---|
375 | tdb->flags |= TDB_CONVERT;
|
---|
376 | tdb_convert(&tdb->header, sizeof(tdb->header));
|
---|
377 | }
|
---|
378 | if (fstat(tdb->fd, &st) == -1)
|
---|
379 | goto fail;
|
---|
380 |
|
---|
381 | if (tdb->header.rwlocks != 0 &&
|
---|
382 | tdb->header.rwlocks != TDB_HASH_RWLOCK_MAGIC) {
|
---|
383 | TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: spinlocks no longer supported\n"));
|
---|
384 | goto fail;
|
---|
385 | }
|
---|
386 |
|
---|
387 | if ((tdb->header.magic1_hash == 0) && (tdb->header.magic2_hash == 0)) {
|
---|
388 | /* older TDB without magic hash references */
|
---|
389 | tdb->hash_fn = tdb_old_hash;
|
---|
390 | } else if (!check_header_hash(tdb, !hash_fn, &magic1, &magic2)) {
|
---|
391 | TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
|
---|
392 | "%s was not created with %s hash function we are using\n"
|
---|
393 | "magic1_hash[0x%08X %s 0x%08X] "
|
---|
394 | "magic2_hash[0x%08X %s 0x%08X]\n",
|
---|
395 | name, hash_alg,
|
---|
396 | tdb->header.magic1_hash,
|
---|
397 | (tdb->header.magic1_hash == magic1) ? "==" : "!=",
|
---|
398 | magic1,
|
---|
399 | tdb->header.magic2_hash,
|
---|
400 | (tdb->header.magic2_hash == magic2) ? "==" : "!=",
|
---|
401 | magic2));
|
---|
402 | errno = EINVAL;
|
---|
403 | goto fail;
|
---|
404 | }
|
---|
405 |
|
---|
406 | /* Is it already in the open list? If so, fail. */
|
---|
407 | if (tdb_already_open(st.st_dev, st.st_ino)) {
|
---|
408 | TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
|
---|
409 | "%s (%d,%d) is already open in this process\n",
|
---|
410 | name, (int)st.st_dev, (int)st.st_ino));
|
---|
411 | errno = EBUSY;
|
---|
412 | goto fail;
|
---|
413 | }
|
---|
414 |
|
---|
415 | tdb->map_size = st.st_size;
|
---|
416 | tdb->device = st.st_dev;
|
---|
417 | tdb->inode = st.st_ino;
|
---|
418 | tdb_mmap(tdb);
|
---|
419 | if (locked) {
|
---|
420 | if (tdb_nest_unlock(tdb, ACTIVE_LOCK, F_WRLCK, false) == -1) {
|
---|
421 | TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
|
---|
422 | "failed to release ACTIVE_LOCK on %s: %s\n",
|
---|
423 | name, strerror(errno)));
|
---|
424 | goto fail;
|
---|
425 | }
|
---|
426 |
|
---|
427 | }
|
---|
428 |
|
---|
429 | /* We always need to do this if the CLEAR_IF_FIRST flag is set, even if
|
---|
430 | we didn't get the initial exclusive lock as we need to let all other
|
---|
431 | users know we're using it. */
|
---|
432 |
|
---|
433 | if (tdb_flags & TDB_CLEAR_IF_FIRST) {
|
---|
434 | /* leave this lock in place to indicate it's in use */
|
---|
435 | if (tdb_nest_lock(tdb, ACTIVE_LOCK, F_RDLCK, TDB_LOCK_WAIT) == -1) {
|
---|
436 | goto fail;
|
---|
437 | }
|
---|
438 | }
|
---|
439 |
|
---|
440 | /* if needed, run recovery */
|
---|
441 | if (tdb_transaction_recover(tdb) == -1) {
|
---|
442 | goto fail;
|
---|
443 | }
|
---|
444 |
|
---|
445 | #ifdef TDB_TRACE
|
---|
446 | {
|
---|
447 | char tracefile[strlen(name) + 32];
|
---|
448 |
|
---|
449 | snprintf(tracefile, sizeof(tracefile),
|
---|
450 | "%s.trace.%li", name, (long)getpid());
|
---|
451 | tdb->tracefd = open(tracefile, O_WRONLY|O_CREAT|O_EXCL, 0600);
|
---|
452 | if (tdb->tracefd >= 0) {
|
---|
453 | tdb_enable_seqnum(tdb);
|
---|
454 | tdb_trace_open(tdb, "tdb_open", hash_size, tdb_flags,
|
---|
455 | open_flags);
|
---|
456 | } else
|
---|
457 | TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to open trace file %s!\n", tracefile));
|
---|
458 | }
|
---|
459 | #endif
|
---|
460 |
|
---|
461 | internal:
|
---|
462 | /* Internal (memory-only) databases skip all the code above to
|
---|
463 | * do with disk files, and resume here by releasing their
|
---|
464 | * open lock and hooking into the active list. */
|
---|
465 | if (tdb_nest_unlock(tdb, OPEN_LOCK, F_WRLCK, false) == -1) {
|
---|
466 | goto fail;
|
---|
467 | }
|
---|
468 | tdb->next = tdbs;
|
---|
469 | tdbs = tdb;
|
---|
470 | return tdb;
|
---|
471 |
|
---|
472 | fail:
|
---|
473 | { int save_errno = errno;
|
---|
474 |
|
---|
475 | if (!tdb)
|
---|
476 | return NULL;
|
---|
477 |
|
---|
478 | #ifdef TDB_TRACE
|
---|
479 | close(tdb->tracefd);
|
---|
480 | #endif
|
---|
481 | if (tdb->map_ptr) {
|
---|
482 | if (tdb->flags & TDB_INTERNAL)
|
---|
483 | SAFE_FREE(tdb->map_ptr);
|
---|
484 | else
|
---|
485 | tdb_munmap(tdb);
|
---|
486 | }
|
---|
487 | if (tdb->fd != -1)
|
---|
488 | if (close(tdb->fd) != 0)
|
---|
489 | TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to close tdb->fd on error!\n"));
|
---|
490 | SAFE_FREE(tdb->lockrecs);
|
---|
491 | SAFE_FREE(tdb->name);
|
---|
492 | SAFE_FREE(tdb);
|
---|
493 | errno = save_errno;
|
---|
494 | return NULL;
|
---|
495 | }
|
---|
496 | }
|
---|
497 |
|
---|
498 | /*
|
---|
499 | * Set the maximum number of dead records per hash chain
|
---|
500 | */
|
---|
501 |
|
---|
502 | _PUBLIC_ void tdb_set_max_dead(struct tdb_context *tdb, int max_dead)
|
---|
503 | {
|
---|
504 | tdb->max_dead_records = max_dead;
|
---|
505 | }
|
---|
506 |
|
---|
507 | /**
|
---|
508 | * Close a database.
|
---|
509 | *
|
---|
510 | * @returns -1 for error; 0 for success.
|
---|
511 | **/
|
---|
512 | _PUBLIC_ int tdb_close(struct tdb_context *tdb)
|
---|
513 | {
|
---|
514 | struct tdb_context **i;
|
---|
515 | int ret = 0;
|
---|
516 |
|
---|
517 | if (tdb->transaction) {
|
---|
518 | tdb_transaction_cancel(tdb);
|
---|
519 | }
|
---|
520 | tdb_trace(tdb, "tdb_close");
|
---|
521 |
|
---|
522 | if (tdb->map_ptr) {
|
---|
523 | if (tdb->flags & TDB_INTERNAL)
|
---|
524 | SAFE_FREE(tdb->map_ptr);
|
---|
525 | else
|
---|
526 | tdb_munmap(tdb);
|
---|
527 | }
|
---|
528 | SAFE_FREE(tdb->name);
|
---|
529 | if (tdb->fd != -1) {
|
---|
530 | ret = close(tdb->fd);
|
---|
531 | tdb->fd = -1;
|
---|
532 | }
|
---|
533 | SAFE_FREE(tdb->lockrecs);
|
---|
534 |
|
---|
535 | /* Remove from contexts list */
|
---|
536 | for (i = &tdbs; *i; i = &(*i)->next) {
|
---|
537 | if (*i == tdb) {
|
---|
538 | *i = tdb->next;
|
---|
539 | break;
|
---|
540 | }
|
---|
541 | }
|
---|
542 |
|
---|
543 | #ifdef TDB_TRACE
|
---|
544 | close(tdb->tracefd);
|
---|
545 | #endif
|
---|
546 | memset(tdb, 0, sizeof(*tdb));
|
---|
547 | SAFE_FREE(tdb);
|
---|
548 |
|
---|
549 | return ret;
|
---|
550 | }
|
---|
551 |
|
---|
552 | /* register a loging function */
|
---|
553 | _PUBLIC_ void tdb_set_logging_function(struct tdb_context *tdb,
|
---|
554 | const struct tdb_logging_context *log_ctx)
|
---|
555 | {
|
---|
556 | tdb->log = *log_ctx;
|
---|
557 | }
|
---|
558 |
|
---|
559 | _PUBLIC_ void *tdb_get_logging_private(struct tdb_context *tdb)
|
---|
560 | {
|
---|
561 | return tdb->log.log_private;
|
---|
562 | }
|
---|
563 |
|
---|
564 | static int tdb_reopen_internal(struct tdb_context *tdb, bool active_lock)
|
---|
565 | {
|
---|
566 | #if !defined(LIBREPLACE_PREAD_NOT_REPLACED) || \
|
---|
567 | !defined(LIBREPLACE_PWRITE_NOT_REPLACED)
|
---|
568 | struct stat st;
|
---|
569 | #endif
|
---|
570 |
|
---|
571 | if (tdb->flags & TDB_INTERNAL) {
|
---|
572 | return 0; /* Nothing to do. */
|
---|
573 | }
|
---|
574 |
|
---|
575 | if (tdb_have_extra_locks(tdb)) {
|
---|
576 | TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed with locks held\n"));
|
---|
577 | goto fail;
|
---|
578 | }
|
---|
579 |
|
---|
580 | if (tdb->transaction != 0) {
|
---|
581 | TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed inside a transaction\n"));
|
---|
582 | goto fail;
|
---|
583 | }
|
---|
584 |
|
---|
585 | /* If we have real pread & pwrite, we can skip reopen. */
|
---|
586 | #if !defined(LIBREPLACE_PREAD_NOT_REPLACED) || \
|
---|
587 | !defined(LIBREPLACE_PWRITE_NOT_REPLACED)
|
---|
588 | if (tdb_munmap(tdb) != 0) {
|
---|
589 | TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: munmap failed (%s)\n", strerror(errno)));
|
---|
590 | goto fail;
|
---|
591 | }
|
---|
592 | if (close(tdb->fd) != 0)
|
---|
593 | TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: WARNING closing tdb->fd failed!\n"));
|
---|
594 | tdb->fd = open(tdb->name, tdb->open_flags & ~(O_CREAT|O_TRUNC), 0);
|
---|
595 | if (tdb->fd == -1) {
|
---|
596 | TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: open failed (%s)\n", strerror(errno)));
|
---|
597 | goto fail;
|
---|
598 | }
|
---|
599 | if (fstat(tdb->fd, &st) != 0) {
|
---|
600 | TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: fstat failed (%s)\n", strerror(errno)));
|
---|
601 | goto fail;
|
---|
602 | }
|
---|
603 | if (st.st_ino != tdb->inode || st.st_dev != tdb->device) {
|
---|
604 | TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: file dev/inode has changed!\n"));
|
---|
605 | goto fail;
|
---|
606 | }
|
---|
607 | tdb_mmap(tdb);
|
---|
608 | #endif /* fake pread or pwrite */
|
---|
609 |
|
---|
610 | /* We may still think we hold the active lock. */
|
---|
611 | tdb->num_lockrecs = 0;
|
---|
612 | SAFE_FREE(tdb->lockrecs);
|
---|
613 |
|
---|
614 | if (active_lock && tdb_nest_lock(tdb, ACTIVE_LOCK, F_RDLCK, TDB_LOCK_WAIT) == -1) {
|
---|
615 | TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: failed to obtain active lock\n"));
|
---|
616 | goto fail;
|
---|
617 | }
|
---|
618 |
|
---|
619 | return 0;
|
---|
620 |
|
---|
621 | fail:
|
---|
622 | tdb_close(tdb);
|
---|
623 | return -1;
|
---|
624 | }
|
---|
625 |
|
---|
626 | /* reopen a tdb - this can be used after a fork to ensure that we have an independent
|
---|
627 | seek pointer from our parent and to re-establish locks */
|
---|
628 | _PUBLIC_ int tdb_reopen(struct tdb_context *tdb)
|
---|
629 | {
|
---|
630 | return tdb_reopen_internal(tdb, tdb->flags & TDB_CLEAR_IF_FIRST);
|
---|
631 | }
|
---|
632 |
|
---|
633 | /* reopen all tdb's */
|
---|
634 | _PUBLIC_ int tdb_reopen_all(int parent_longlived)
|
---|
635 | {
|
---|
636 | struct tdb_context *tdb;
|
---|
637 |
|
---|
638 | for (tdb=tdbs; tdb; tdb = tdb->next) {
|
---|
639 | bool active_lock = (tdb->flags & TDB_CLEAR_IF_FIRST);
|
---|
640 |
|
---|
641 | /*
|
---|
642 | * If the parent is longlived (ie. a
|
---|
643 | * parent daemon architecture), we know
|
---|
644 | * it will keep it's active lock on a
|
---|
645 | * tdb opened with CLEAR_IF_FIRST. Thus
|
---|
646 | * for child processes we don't have to
|
---|
647 | * add an active lock. This is essential
|
---|
648 | * to improve performance on systems that
|
---|
649 | * keep POSIX locks as a non-scalable data
|
---|
650 | * structure in the kernel.
|
---|
651 | */
|
---|
652 | if (parent_longlived) {
|
---|
653 | /* Ensure no clear-if-first. */
|
---|
654 | active_lock = false;
|
---|
655 | }
|
---|
656 |
|
---|
657 | if (tdb_reopen_internal(tdb, active_lock) != 0)
|
---|
658 | return -1;
|
---|
659 | }
|
---|
660 |
|
---|
661 | return 0;
|
---|
662 | }
|
---|