Line data Source code
1 : #ifndef TDB_PRIVATE_H
2 : #define TDB_PRIVATE_H
3 : /*
4 : Unix SMB/CIFS implementation.
5 :
6 : trivial database library - private includes
7 :
8 : Copyright (C) Andrew Tridgell 2005
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 "replace.h"
29 : #include "system/filesys.h"
30 : #include "system/time.h"
31 : #include "system/shmem.h"
32 : #include "system/select.h"
33 : #include "system/wait.h"
34 : #include "tdb.h"
35 :
36 : /* #define TDB_TRACE 1 */
37 : #ifndef HAVE_GETPAGESIZE
38 : #define getpagesize() 0x2000
39 : #endif
40 :
41 : typedef uint32_t tdb_len_t;
42 : typedef uint32_t tdb_off_t;
43 :
44 : #ifndef offsetof
45 : #define offsetof(t,f) ((unsigned int)&((t *)0)->f)
46 : #endif
47 :
48 : #define TDB_MAGIC_FOOD "TDB file\n"
49 : #define TDB_VERSION (0x26011967 + 6)
50 : #define TDB_MAGIC (0x26011999U)
51 : #define TDB_FREE_MAGIC (~TDB_MAGIC)
52 : #define TDB_DEAD_MAGIC (0xFEE1DEAD)
53 : #define TDB_RECOVERY_MAGIC (0xf53bc0e7U)
54 : #define TDB_RECOVERY_INVALID_MAGIC (0x0)
55 : #define TDB_HASH_RWLOCK_MAGIC (0xbad1a51U)
56 : #define TDB_FEATURE_FLAG_MAGIC (0xbad1a52U)
57 : #define TDB_ALIGNMENT 4
58 : #define DEFAULT_HASH_SIZE 131
59 : #define FREELIST_TOP (sizeof(struct tdb_header))
60 : #define TDB_ALIGN(x,a) (((x) + (a)-1) & ~((a)-1))
61 : #define TDB_BYTEREV(x) (((((x)&0xff)<<24)|((x)&0xFF00)<<8)|(((x)>>8)&0xFF00)|((x)>>24))
62 : #define TDB_DEAD(r) ((r)->magic == TDB_DEAD_MAGIC)
63 : #define TDB_BAD_MAGIC(r) ((r)->magic != TDB_MAGIC && !TDB_DEAD(r))
64 : #define TDB_HASH_TOP(hash) (FREELIST_TOP + (BUCKET(hash)+1)*sizeof(tdb_off_t))
65 : #define TDB_HASHTABLE_SIZE(tdb) ((tdb->hash_size+1)*sizeof(tdb_off_t))
66 : #define TDB_DATA_START(hash_size) (TDB_HASH_TOP(hash_size-1) + sizeof(tdb_off_t))
67 : #define TDB_RECOVERY_HEAD offsetof(struct tdb_header, recovery_start)
68 : #define TDB_SEQNUM_OFS offsetof(struct tdb_header, sequence_number)
69 : #define TDB_PAD_BYTE 0x42
70 : #define TDB_PAD_U32 0x42424242
71 :
72 : #define TDB_FEATURE_FLAG_MUTEX 0x00000001
73 :
74 : #define TDB_SUPPORTED_FEATURE_FLAGS ( \
75 : TDB_FEATURE_FLAG_MUTEX | \
76 : 0)
77 :
78 : /* NB assumes there is a local variable called "tdb" that is the
79 : * current context, also takes doubly-parenthesized print-style
80 : * argument. */
81 : #define TDB_LOG(x) tdb->log.log_fn x
82 :
83 : #ifdef TDB_TRACE
84 : void tdb_trace(struct tdb_context *tdb, const char *op);
85 : void tdb_trace_seqnum(struct tdb_context *tdb, uint32_t seqnum, const char *op);
86 : void tdb_trace_open(struct tdb_context *tdb, const char *op,
87 : unsigned hash_size, unsigned tdb_flags, unsigned open_flags);
88 : void tdb_trace_ret(struct tdb_context *tdb, const char *op, int ret);
89 : void tdb_trace_retrec(struct tdb_context *tdb, const char *op, TDB_DATA ret);
90 : void tdb_trace_1rec(struct tdb_context *tdb, const char *op,
91 : TDB_DATA rec);
92 : void tdb_trace_1rec_ret(struct tdb_context *tdb, const char *op,
93 : TDB_DATA rec, int ret);
94 : void tdb_trace_1rec_retrec(struct tdb_context *tdb, const char *op,
95 : TDB_DATA rec, TDB_DATA ret);
96 : void tdb_trace_2rec_flag_ret(struct tdb_context *tdb, const char *op,
97 : TDB_DATA rec1, TDB_DATA rec2, unsigned flag,
98 : int ret);
99 : void tdb_trace_1plusn_rec_flag_ret(struct tdb_context *tdb, const char *op,
100 : TDB_DATA rec,
101 : const TDB_DATA *recs, int num_recs,
102 : unsigned flag, int ret);
103 : void tdb_trace_2rec_retrec(struct tdb_context *tdb, const char *op,
104 : TDB_DATA rec1, TDB_DATA rec2, TDB_DATA ret);
105 : #else
106 : #define tdb_trace(tdb, op)
107 : #define tdb_trace_seqnum(tdb, seqnum, op)
108 : #define tdb_trace_open(tdb, op, hash_size, tdb_flags, open_flags)
109 : #define tdb_trace_ret(tdb, op, ret)
110 : #define tdb_trace_retrec(tdb, op, ret)
111 : #define tdb_trace_1rec(tdb, op, rec)
112 : #define tdb_trace_1rec_ret(tdb, op, rec, ret)
113 : #define tdb_trace_1rec_retrec(tdb, op, rec, ret)
114 : #define tdb_trace_2rec_flag_ret(tdb, op, rec1, rec2, flag, ret)
115 : #define tdb_trace_1plusn_rec_flag_ret(tdb, op, rec, recs, num_recs, flag, ret);
116 : #define tdb_trace_2rec_retrec(tdb, op, rec1, rec2, ret)
117 : #endif /* !TDB_TRACE */
118 :
119 : /* lock offsets */
120 : #define OPEN_LOCK 0
121 : #define ACTIVE_LOCK 4
122 : #define TRANSACTION_LOCK 8
123 :
124 : /* free memory if the pointer is valid and zero the pointer */
125 : #ifndef SAFE_FREE
126 : #define SAFE_FREE(x) do { if ((x) != NULL) {free(x); (x)=NULL;} } while(0)
127 : #endif
128 :
129 : /*
130 : * Note: the BUCKET macro is broken as it returns an unexpected result when
131 : * called as BUCKET(-1) for the freelist:
132 : *
133 : * -1 is sign converted to an unsigned int 4294967295 and then the modulo
134 : * tdb->hashtable_size is computed. So with a hashtable_size of 10 the result
135 : * is
136 : *
137 : * 4294967295 % hashtable_size = 5.
138 : *
139 : * where it should be -1 (C uses symmetric modulo).
140 : *
141 : * As all callers will lock the same wrong list consistently locking is still
142 : * consistent. We can not change this without an incompatible on-disk format
143 : * change, otherwise different tdb versions would use incompatible locking.
144 : */
145 : #define BUCKET(hash) ((hash) % tdb->hash_size)
146 :
147 : #define DOCONV() (tdb->flags & TDB_CONVERT)
148 : #define CONVERT(x) (DOCONV() ? tdb_convert(&x, sizeof(x)) : &x)
149 :
150 :
151 : /* the body of the database is made of one tdb_record for the free space
152 : plus a separate data list for each hash value */
153 : struct tdb_record {
154 : tdb_off_t next; /* offset of the next record in the list */
155 : tdb_len_t rec_len; /* total byte length of record */
156 : tdb_len_t key_len; /* byte length of key */
157 : tdb_len_t data_len; /* byte length of data */
158 : uint32_t full_hash; /* the full 32 bit hash of the key */
159 : uint32_t magic; /* try to catch errors */
160 : /* the following union is implied:
161 : union {
162 : char record[rec_len];
163 : struct {
164 : char key[key_len];
165 : char data[data_len];
166 : }
167 : uint32_t totalsize; (tailer)
168 : }
169 : */
170 : };
171 :
172 :
173 : /* this is stored at the front of every database */
174 : struct tdb_header {
175 : char magic_food[32]; /* for /etc/magic */
176 : uint32_t version; /* version of the code */
177 : uint32_t hash_size; /* number of hash entries */
178 : tdb_off_t rwlocks; /* obsolete - kept to detect old formats */
179 : tdb_off_t recovery_start; /* offset of transaction recovery region */
180 : tdb_off_t sequence_number; /* used when TDB_SEQNUM is set */
181 : uint32_t magic1_hash; /* hash of TDB_MAGIC_FOOD. */
182 : uint32_t magic2_hash; /* hash of TDB_MAGIC. */
183 : uint32_t feature_flags;
184 : tdb_len_t mutex_size; /* set if TDB_FEATURE_FLAG_MUTEX is set */
185 : tdb_off_t reserved[25];
186 : };
187 :
188 : struct tdb_lock_type {
189 : uint32_t off;
190 : uint32_t count;
191 : uint32_t ltype;
192 : };
193 :
194 : struct tdb_chainwalk_ctx {
195 : tdb_off_t slow_ptr;
196 : bool slow_chase;
197 : };
198 :
199 : struct tdb_traverse_lock {
200 : struct tdb_traverse_lock *next;
201 : uint32_t off;
202 : uint32_t list;
203 : int lock_rw;
204 : };
205 :
206 : void tdb_chainwalk_init(struct tdb_chainwalk_ctx *ctx, tdb_off_t ptr);
207 : bool tdb_chainwalk_check(struct tdb_context *tdb,
208 : struct tdb_chainwalk_ctx *ctx,
209 : tdb_off_t next_ptr);
210 :
211 : enum tdb_lock_flags {
212 : /* WAIT == F_SETLKW, NOWAIT == F_SETLK */
213 : TDB_LOCK_NOWAIT = 0,
214 : TDB_LOCK_WAIT = 1,
215 : /* If set, don't log an error on failure. */
216 : TDB_LOCK_PROBE = 2,
217 : /* If set, don't actually lock at all. */
218 : TDB_LOCK_MARK_ONLY = 4,
219 : };
220 :
221 : struct tdb_methods {
222 : int (*tdb_read)(struct tdb_context *, tdb_off_t , void *, tdb_len_t , int );
223 : int (*tdb_write)(struct tdb_context *, tdb_off_t, const void *, tdb_len_t);
224 : void (*next_hash_chain)(struct tdb_context *, uint32_t *);
225 : int (*tdb_oob)(struct tdb_context *, tdb_off_t , tdb_len_t, int );
226 : int (*tdb_expand_file)(struct tdb_context *, tdb_off_t , tdb_off_t );
227 : };
228 :
229 : struct tdb_mutexes;
230 :
231 : struct tdb_context {
232 : char *name; /* the name of the database */
233 : void *map_ptr; /* where it is currently mapped */
234 : int fd; /* open file descriptor for the database */
235 : tdb_len_t map_size; /* how much space has been mapped */
236 : int read_only; /* opened read-only */
237 : int traverse_read; /* read-only traversal */
238 : int traverse_write; /* read-write traversal */
239 : struct tdb_lock_type allrecord_lock; /* .offset == upgradable */
240 : int num_lockrecs;
241 : struct tdb_lock_type *lockrecs; /* only real locks, all with count>0 */
242 : int lockrecs_array_length;
243 :
244 : tdb_off_t hdr_ofs; /* this is 0 or header.mutex_size */
245 : struct tdb_mutexes *mutexes; /* mmap of the mutex area */
246 :
247 : enum TDB_ERROR ecode; /* error code for last tdb error */
248 : uint32_t hash_size;
249 : uint32_t feature_flags;
250 : uint32_t flags; /* the flags passed to tdb_open */
251 : struct tdb_traverse_lock travlocks; /* current traversal locks */
252 : struct tdb_context *next; /* all tdbs to avoid multiple opens */
253 : dev_t device; /* uniquely identifies this tdb */
254 : ino_t inode; /* uniquely identifies this tdb */
255 : struct tdb_logging_context log;
256 : unsigned int (*hash_fn)(TDB_DATA *key);
257 : int open_flags; /* flags used in the open - needed by reopen */
258 : const struct tdb_methods *methods;
259 : struct tdb_transaction *transaction;
260 : int page_size;
261 : int max_dead_records;
262 : #ifdef TDB_TRACE
263 : int tracefd;
264 : #endif
265 : volatile sig_atomic_t *interrupt_sig_ptr;
266 : };
267 :
268 :
269 : /*
270 : internal prototypes
271 : */
272 : int tdb_munmap(struct tdb_context *tdb);
273 : int tdb_mmap(struct tdb_context *tdb);
274 : int tdb_lock(struct tdb_context *tdb, int list, int ltype);
275 : int tdb_lock_nonblock(struct tdb_context *tdb, int list, int ltype);
276 : int tdb_nest_lock(struct tdb_context *tdb, uint32_t offset, int ltype,
277 : enum tdb_lock_flags flags);
278 : int tdb_nest_unlock(struct tdb_context *tdb, uint32_t offset, int ltype,
279 : bool mark_lock);
280 : int tdb_unlock(struct tdb_context *tdb, int list, int ltype);
281 : int tdb_brlock(struct tdb_context *tdb,
282 : int rw_type, tdb_off_t offset, size_t len,
283 : enum tdb_lock_flags flags);
284 : int tdb_brunlock(struct tdb_context *tdb,
285 : int rw_type, tdb_off_t offset, size_t len);
286 : bool tdb_have_extra_locks(struct tdb_context *tdb);
287 : void tdb_release_transaction_locks(struct tdb_context *tdb);
288 : int tdb_transaction_lock(struct tdb_context *tdb, int ltype,
289 : enum tdb_lock_flags lockflags);
290 : int tdb_transaction_unlock(struct tdb_context *tdb, int ltype);
291 : int tdb_recovery_area(struct tdb_context *tdb,
292 : const struct tdb_methods *methods,
293 : tdb_off_t *recovery_offset,
294 : struct tdb_record *rec);
295 : int tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
296 : enum tdb_lock_flags flags, bool upgradable);
297 : int tdb_allrecord_unlock(struct tdb_context *tdb, int ltype, bool mark_lock);
298 : int tdb_allrecord_upgrade(struct tdb_context *tdb);
299 : int tdb_write_lock_record(struct tdb_context *tdb, tdb_off_t off);
300 : int tdb_write_unlock_record(struct tdb_context *tdb, tdb_off_t off);
301 : int tdb_ofs_read(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d);
302 : int tdb_ofs_write(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d);
303 : void *tdb_convert(void *buf, uint32_t size);
304 : int tdb_free(struct tdb_context *tdb, tdb_off_t offset, struct tdb_record *rec);
305 : tdb_off_t tdb_allocate(struct tdb_context *tdb, int hash, tdb_len_t length,
306 : struct tdb_record *rec);
307 :
308 : int _tdb_oob(struct tdb_context *tdb, tdb_off_t off, tdb_len_t len, int probe);
309 :
310 25589685830 : static inline int tdb_oob(
311 : struct tdb_context *tdb, tdb_off_t off, tdb_len_t len, int probe)
312 : {
313 24788295168 : if (likely((off + len >= off) && (off + len <= tdb->map_size))) {
314 24056635520 : return 0;
315 : }
316 7061750 : return _tdb_oob(tdb, off, len, probe);
317 : }
318 :
319 :
320 : int tdb_ofs_read(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d);
321 : int tdb_ofs_write(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d);
322 : int tdb_lock_record(struct tdb_context *tdb, tdb_off_t off);
323 : int tdb_unlock_record(struct tdb_context *tdb, tdb_off_t off);
324 : bool tdb_needs_recovery(struct tdb_context *tdb);
325 : int tdb_rec_read(struct tdb_context *tdb, tdb_off_t offset, struct tdb_record *rec);
326 : int tdb_rec_write(struct tdb_context *tdb, tdb_off_t offset, struct tdb_record *rec);
327 : unsigned char *tdb_alloc_read(struct tdb_context *tdb, tdb_off_t offset, tdb_len_t len);
328 : int tdb_parse_data(struct tdb_context *tdb, TDB_DATA key,
329 : tdb_off_t offset, tdb_len_t len,
330 : int (*parser)(TDB_DATA key, TDB_DATA data,
331 : void *private_data),
332 : void *private_data);
333 : tdb_off_t tdb_find_lock_hash(struct tdb_context *tdb, TDB_DATA key, uint32_t hash, int locktype,
334 : struct tdb_record *rec);
335 : tdb_off_t tdb_find_dead(struct tdb_context *tdb, uint32_t hash,
336 : struct tdb_record *r, tdb_len_t length,
337 : tdb_off_t *p_last_ptr);
338 : int tdb_trim_dead(struct tdb_context *tdb, uint32_t hash);
339 : void tdb_io_init(struct tdb_context *tdb);
340 : int tdb_expand(struct tdb_context *tdb, tdb_off_t size);
341 : tdb_off_t tdb_expand_adjust(tdb_off_t map_size, tdb_off_t size, int page_size);
342 : int tdb_rec_free_read(struct tdb_context *tdb, tdb_off_t off,
343 : struct tdb_record *rec);
344 : bool tdb_write_all(int fd, const void *buf, size_t count);
345 : int tdb_transaction_recover(struct tdb_context *tdb);
346 : void tdb_header_hash(struct tdb_context *tdb,
347 : uint32_t *magic1_hash, uint32_t *magic2_hash);
348 : unsigned int tdb_old_hash(TDB_DATA *key);
349 : size_t tdb_dead_space(struct tdb_context *tdb, tdb_off_t off);
350 : bool tdb_add_off_t(tdb_off_t a, tdb_off_t b, tdb_off_t *pret);
351 :
352 : /* tdb_off_t and tdb_len_t right now are both uint32_t */
353 : #define tdb_add_len_t tdb_add_off_t
354 :
355 : size_t tdb_mutex_size(struct tdb_context *tdb);
356 : bool tdb_have_mutexes(struct tdb_context *tdb);
357 : int tdb_mutex_init(struct tdb_context *tdb);
358 : int tdb_mutex_mmap(struct tdb_context *tdb);
359 : int tdb_mutex_munmap(struct tdb_context *tdb);
360 : bool tdb_mutex_lock(struct tdb_context *tdb, int rw, off_t off, off_t len,
361 : bool waitflag, int *pret);
362 : bool tdb_mutex_unlock(struct tdb_context *tdb, int rw, off_t off, off_t len,
363 : int *pret);
364 : int tdb_mutex_allrecord_lock(struct tdb_context *tdb, int ltype,
365 : enum tdb_lock_flags flags);
366 : int tdb_mutex_allrecord_unlock(struct tdb_context *tdb);
367 : int tdb_mutex_allrecord_upgrade(struct tdb_context *tdb);
368 : void tdb_mutex_allrecord_downgrade(struct tdb_context *tdb);
369 :
370 : #endif /* TDB_PRIVATE_H */
|
