loglevel adjustments.
[paraslash.git] / osl.c
1 /*
2 * Copyright (C) 2007 Andre Noll <maan@systemlinux.org>
3 *
4 * Licensed under the GPL v2. For licencing details see COPYING.
5 */
6
7 /** \file osl.c Object storage layer functions. */
8 #include "para.h"
9 #include "error.h"
10 #include "list.h"
11 #include "osl_core.h"
12 #include <dirent.h> /* readdir() */
13 #include <assert.h>
14
15 //#define FMT_OFF_T "%li"
16
17
18 /**
19 * A wrapper for lseek(2).
20 *
21 * \param fd The filedescriptor whose offset is to be to repositioned.
22 * \param offset A value-result parameter.
23 * \param whence Usual repositioning directive.
24 *
25 * Reposition the offset of the file descriptor \a fd to the argument \a offset
26 * according to the directive \a whence. Upon successful return, \a offset
27 * contains the resulting offset location as measured in bytes from the
28 * beginning of the file.
29 *
30 * \return Positive on success. Otherwise, the function returns \p -E_LSEEK.
31 *
32 * \sa lseek(2).
33 */
34 int para_lseek(int fd, off_t *offset, int whence)
35 {
36 *offset = lseek(fd, *offset, whence);
37 int ret = -E_LSEEK;
38 if (*offset == -1)
39 return ret;
40 return 1;
41 }
42
43 /**
44 * Waraper for the write system call.
45 *
46 * \param fd The file descriptor to write to.
47 * \param buf The buffer to write.
48 * \param size The length of \a buf in bytes.
49 *
50 * This function writes out the given bufffer and retries if an interrupt
51 * occured during the write.
52 *
53 * \return On success, the number of bytes written is returned, otherwise, the
54 * function returns \p -E_WRITE.
55 *
56 * \sa write(2).
57 */
58 ssize_t para_write(int fd, const void *buf, size_t size)
59 {
60 ssize_t ret;
61
62 for (;;) {
63 ret = write(fd, buf, size);
64 if ((ret < 0) && (errno == EAGAIN || errno == EINTR))
65 continue;
66 return ret >= 0? ret : -E_WRITE;
67 }
68 }
69
70 /**
71 * Write the whole buffer to a file descriptor.
72 *
73 * \param fd The file descriptor to write to.
74 * \param buf The buffer to write.
75 * \param size The length of \a buf in bytes.
76 *
77 * This function writes the given buffer and continues on short writes and
78 * when interrupted by a signal.
79 *
80 * \return Positive on success, negative on errors. Possible errors: any
81 * errors returned by para_write().
82 *
83 * \sa para_write().
84 */
85 ssize_t para_write_all(int fd, const void *buf, size_t size)
86 {
87 // PARA_DEBUG_LOG("writing %zu bytes\n", size);
88 const char *b = buf;
89 while (size) {
90 ssize_t ret = para_write(fd, b, size);
91 // PARA_DEBUG_LOG("ret: %zd\n", ret);
92 if (ret < 0)
93 return ret;
94 b += ret;
95 size -= ret;
96 }
97 return 1;
98 }
99 /**
100 * Wrapper for the open(2) system call.
101 *
102 * \param path The filename.
103 * \param flags The usual open(2) flags.
104 * \param mode Specifies the permissions to use.
105 *
106 * The mode parameter must be specified when O_CREAT is in the flags, and is ignored
107 * otherwise.
108 *
109 * \return Positive on success, negative on errors. Possible errors: \p
110 * E_EXIST, \p E_ISDIR, \p E_NOENT, \p E_OSL_PERM.
111 *
112 * \sa open(2).
113 */
114 int para_open(const char *path, int flags, mode_t mode)
115 {
116 PARA_DEBUG_LOG("opening %s\n", path);
117 int ret = open(path, flags, mode);
118
119 if (ret >= 0)
120 return ret;
121 switch (errno) {
122 case EEXIST:
123 ret = -E_EXIST;
124 break;
125 case EISDIR:
126 ret = -E_ISDIR;
127 break;
128 case ENOENT:
129 ret = -E_NOENT;
130 break;
131 case EPERM:
132 ret = -E_OSL_PERM;
133 break;
134 };
135 PARA_ERROR_LOG("failed to open %s: %s\n", path, strerror(errno));
136 return ret;
137 }
138
139 /**
140 * Open a file, write the given buffer and close the file.
141 *
142 * \param filename Full path to the file to open.
143 * \param buf The buffer to write to the file.
144 * \param size The size of \a buf.
145 *
146 * \return Positive on success, negative on errors. Possible errors include:
147 * any errors from para_open() or para_write().
148 *
149 * \sa para_open(), para_write().
150 */
151 int para_write_file(const char *filename, const void *buf, size_t size)
152 {
153 int ret, fd;
154
155 ret = para_open(filename, O_WRONLY | O_CREAT | O_EXCL, 0644);
156 if (ret < 0)
157 return ret;
158 fd = ret;
159 ret = para_write_all(fd, buf, size);
160 if (ret < 0)
161 goto out;
162 ret = 1;
163 out:
164 close(fd);
165 return ret;
166 }
167
168 static int append_file(const char *filename, char *header, size_t header_size,
169 char *data, size_t data_size, uint32_t *new_pos)
170 {
171 int ret, fd;
172
173 // PARA_DEBUG_LOG("appending %zu + %zu bytes\n", header_size, data_size);
174 ret = para_open(filename, O_WRONLY | O_CREAT | O_APPEND, 0644);
175 if (ret < 0)
176 return ret;
177 fd = ret;
178 if (header && header_size) {
179 ret = para_write_all(fd, header, header_size);
180 if (ret < 0)
181 goto out;
182 }
183 ret = para_write_all(fd, data, data_size);
184 if (ret < 0)
185 goto out;
186 if (new_pos) {
187 off_t offset = 0;
188 ret = para_lseek(fd, &offset, SEEK_END);
189 if (ret < 0)
190 goto out;
191 // PARA_DEBUG_LOG("new file size: " FMT_OFF_T "\n", offset);
192 *new_pos = offset;
193 }
194 ret = 1;
195 out:
196 close(fd);
197 return ret;
198 }
199
200 /**
201 * Map a file into memory.
202 *
203 * \param path Name of the regular file to map.
204 * \param open_mode Either \p O_RDONLY or \p O_RDWR.
205 * \param obj On success, the mapping is returned here.
206 *
207 * \return Positive on success, negative on errors. Possible errors include: \p
208 * E_FSTAT, any errors returned by para_open(), \p E_EMPTY, \p E_MMAP.
209 *
210 * \sa para_open(), mmap(2).
211 */
212 int mmap_full_file(const char *path, int open_mode, struct osl_object *obj)
213 {
214 int fd, ret, mmap_prot, mmap_flags;
215 struct stat file_status;
216
217 if (open_mode == O_RDONLY) {
218 mmap_prot = PROT_READ;
219 mmap_flags = MAP_PRIVATE;
220 } else {
221 mmap_prot = PROT_READ | PROT_WRITE;
222 mmap_flags = MAP_SHARED;
223 }
224 ret = para_open(path, open_mode, 0);
225 if (ret < 0)
226 return ret;
227 fd = ret;
228 ret = -E_FSTAT;
229 if (fstat(fd, &file_status) < 0)
230 goto out;
231 obj->size = file_status.st_size;
232 ret = -E_EMPTY;
233 PARA_DEBUG_LOG("%s: size %zu\n", path, obj->size);
234 if (!obj->size)
235 goto out;
236 obj->data = mmap(NULL, obj->size, mmap_prot, mmap_flags, fd, 0);
237 if (obj->data == MAP_FAILED) {
238 obj->data = NULL;
239 ret = -E_MMAP;
240 goto out;
241 }
242 ret = 1;
243 out:
244 close(fd);
245 return ret;
246 }
247
248 /**
249 * Traverse the given directory recursively.
250 *
251 * \param dirname The directory to traverse.
252 * \param func The function to call for each entry.
253 * \param private_data Pointer to an arbitrary data structure.
254 *
255 * For each regular file in \a dirname, the supplied function \a func is
256 * called. The full path of the regular file and the \a private_data pointer
257 * are passed to \a func.
258 *
259 * \return On success, 1 is returned. Otherwise, this function returns a
260 * negative value which indicates the kind of the error.
261 */
262 int for_each_file_in_dir(const char *dirname,
263 int (*func)(const char *, const void *), const void *private_data)
264 {
265 DIR *dir = NULL;
266 struct dirent *entry;
267 /*
268 * Opening the current directory (".") and calling fchdir() to return
269 * is usually faster and more reliable than saving cwd in some buffer
270 * and calling chdir() afterwards (see man 3 getcwd).
271 */
272 int cwd_fd = open(".", O_RDONLY);
273 struct stat s;
274 int ret = -1;
275
276 // PARA_DEBUG_LOG("dirname: %s\n", dirname);
277 if (cwd_fd < 0)
278 return -E_OSL_GETCWD;
279 ret = -E_OSL_CHDIR;
280 if (chdir(dirname) < 0)
281 goto out;
282 ret = -E_OSL_OPENDIR;
283 dir = opendir(".");
284 if (!dir)
285 goto out;
286 /* scan cwd recursively */
287 while ((entry = readdir(dir))) {
288 mode_t m;
289 char *tmp;
290
291 if (!strcmp(entry->d_name, "."))
292 continue;
293 if (!strcmp(entry->d_name, ".."))
294 continue;
295 ret = -E_OSL_LSTAT;
296 if (lstat(entry->d_name, &s) == -1)
297 continue;
298 m = s.st_mode;
299 if (!S_ISREG(m) && !S_ISDIR(m))
300 continue;
301 tmp = make_message("%s/%s", dirname, entry->d_name);
302 if (!S_ISDIR(m)) {
303 ret = func(tmp, private_data);
304 free(tmp);
305 if (ret < 0)
306 goto out;
307 continue;
308 }
309 /* directory */
310 ret = for_each_file_in_dir(tmp, func, private_data);
311 free(tmp);
312 if (ret < 0)
313 goto out;
314 }
315 ret = 1;
316 out:
317 if (dir)
318 closedir(dir);
319 if (fchdir(cwd_fd) < 0 && ret >= 0)
320 ret = -E_OSL_CHDIR;
321 close(cwd_fd);
322 return ret;
323 }
324
325 int para_mkdir(const char *path, mode_t mode)
326 {
327 if (!mkdir(path, mode))
328 return 1;
329 if (errno == EEXIST)
330 return -E_EXIST;
331 if (errno == ENOSPC)
332 return -E_NOSPC;
333 if (errno == ENOTDIR)
334 return -E_NOTDIR;
335 if (errno == EPERM)
336 return E_OSL_PERM;
337 return -E_MKDIR;
338 }
339
340 static int verify_basename(const char *name)
341 {
342 if (!name)
343 return -E_BAD_BASENAME;
344 if (!*name)
345 return -E_BAD_BASENAME;
346 if (strchr(name, '/'))
347 return -E_BAD_BASENAME;
348 if (!strcmp(name, ".."))
349 return -E_BAD_BASENAME;
350 if (!strcmp(name, "."))
351 return -E_BAD_BASENAME;
352 return 1;
353 }
354
355 /**
356 * Compare two osl objects pointing to unsigned integers of 32 bit size.
357 *
358 * \param obj1 Pointer to the first integer.
359 * \param obj2 Pointer to the second integer.
360 *
361 * \return The values required for an osl compare function.
362 *
363 * \sa osl_compare_func, osl_hash_compare().
364 */
365 int uint32_compare(const struct osl_object *obj1, const struct osl_object *obj2)
366 {
367 uint32_t d1 = read_u32((const char *)obj1->data);
368 uint32_t d2 = read_u32((const char *)obj2->data);
369
370 if (d1 < d2)
371 return 1;
372 if (d1 > d2)
373 return -1;
374 return 0;
375 }
376
377 /**
378 * Compare two osl objects pointing to hash values.
379 *
380 * \param obj1 Pointer to the first hash object.
381 * \param obj2 Pointer to the second hash object.
382 *
383 * \return The values required for an osl compare function.
384 *
385 * \sa osl_compare_func, uint32_compare().
386 */
387 int osl_hash_compare(const struct osl_object *obj1, const struct osl_object *obj2)
388 {
389 return hash_compare((HASH_TYPE *)obj1->data, (HASH_TYPE *)obj2->data);
390 }
391
392 static char *disk_storage_dirname(const struct osl_table *t, unsigned col_num,
393 const char *ds_name)
394 {
395 char *dirname, *column_name = column_filename(t, col_num);
396
397 if (!(t->desc->flags & OSL_LARGE_TABLE))
398 return column_name;
399 dirname = make_message("%s/%.2s", column_name, ds_name);
400 free(column_name);
401 return dirname;
402 }
403
404 static char *disk_storage_name_of_object(const struct osl_table *t,
405 const struct osl_object *obj)
406 {
407 HASH_TYPE hash[HASH_SIZE];
408 hash_object(obj, hash);
409 return disk_storage_name_of_hash(t, hash);
410 }
411
412 static int disk_storage_name_of_row(const struct osl_table *t,
413 const struct osl_row *row, char **name)
414 {
415 struct osl_object obj;
416 int ret = osl_get_object(t, row, t->disk_storage_name_column, &obj);
417
418 if (ret < 0)
419 return ret;
420 *name = disk_storage_name_of_object(t, &obj);
421 return 1;
422 }
423
424 static void column_name_hash(const char *col_name, HASH_TYPE *hash)
425 {
426 return hash_function(col_name, strlen(col_name), hash);
427 }
428
429 static int init_column_descriptions(struct osl_table *t)
430 {
431 int i, j, ret;
432 const struct osl_column_description *cd;
433
434 ret = -E_BAD_TABLE_DESC;
435 ret = verify_basename(t->desc->name);
436 if (ret < 0)
437 goto err;
438 ret = -E_BAD_DB_DIR;
439 if (!t->desc->dir)
440 goto err;
441 /* the size of the index header without column descriptions */
442 t->index_header_size = IDX_COLUMN_DESCRIPTIONS;
443 FOR_EACH_COLUMN(i, t->desc, cd) {
444 struct osl_column *col = t->columns + i;
445 if (cd->storage_flags & OSL_RBTREE) {
446 if (!cd->compare_function)
447 return -E_NO_COMPARE_FUNC;
448 }
449 if (cd->storage_type == OSL_NO_STORAGE)
450 continue;
451 ret = -E_NO_COLUMN_NAME;
452 if (!cd->name || !cd->name[0])
453 goto err;
454 ret = verify_basename(cd->name);
455 if (ret < 0)
456 goto err;
457 t->index_header_size += index_column_description_size(cd->name);
458 column_name_hash(cd->name, col->name_hash);
459 ret = -E_DUPLICATE_COL_NAME;
460 for (j = i + 1; j < t->desc->num_columns; j++) {
461 const char *name2 = get_column_description(t->desc,
462 j)->name;
463 if (cd->name && name2 && !strcmp(cd->name, name2))
464 goto err;
465 }
466 }
467 return 1;
468 err:
469 return ret;
470 }
471
472 /**
473 * Initialize a struct table from given table description.
474 *
475 * \param desc The description of the osl table.
476 * \param table_ptr Result is returned here.
477 *
478 * This function performs several sanity checks on \p desc and returns if any
479 * of these tests fail. On success, a struct \p osl_table is allocated and
480 * initialized with data derived from \p desc.
481 *
482 * \return Positive on success, negative on errors. Possible errors include: \p
483 * E_BAD_TABLE_DESC, \p E_NO_COLUMN_DESC, \p E_NO_COLUMNS, \p
484 * E_BAD_STORAGE_TYPE, \p E_BAD_STORAGE_FLAGS, \p E_BAD_STORAGE_SIZE, \p
485 * E_NO_UNIQUE_RBTREE_COLUMN, \p E_NO_RBTREE_COL.
486 *
487 * \sa struct osl_table.
488 */
489 int init_table_structure(const struct osl_table_description *desc,
490 struct osl_table **table_ptr)
491 {
492 const struct osl_column_description *cd;
493 struct osl_table *t = para_calloc(sizeof(*t));
494 int i, ret = -E_BAD_TABLE_DESC, have_disk_storage_name_column = 0;
495
496 if (!desc)
497 goto err;
498 PARA_DEBUG_LOG("creating table structure for '%s' from table "
499 "description\n", desc->name);
500 ret = -E_NO_COLUMN_DESC;
501 if (!desc->column_descriptions)
502 goto err;
503 ret = -E_NO_COLUMNS;
504 if (!desc->num_columns)
505 goto err;
506 t->columns = para_calloc(desc->num_columns * sizeof(struct osl_column));
507 t->desc = desc;
508 FOR_EACH_COLUMN(i, t->desc, cd) {
509 enum osl_storage_type st = cd->storage_type;
510 enum osl_storage_flags sf = cd->storage_flags;
511 struct osl_column *col = &t->columns[i];
512
513 ret = -E_BAD_STORAGE_TYPE;
514 if (st != OSL_MAPPED_STORAGE && st != OSL_DISK_STORAGE
515 && st != OSL_NO_STORAGE)
516 goto err;
517 ret = -E_BAD_STORAGE_FLAGS;
518 if (st == OSL_DISK_STORAGE && sf & OSL_RBTREE)
519 goto err;
520 ret = -E_BAD_STORAGE_SIZE;
521 if (sf & OSL_FIXED_SIZE && !cd->data_size)
522 goto err;
523 switch (st) {
524 case OSL_DISK_STORAGE:
525 t->num_disk_storage_columns++;
526 break;
527 case OSL_MAPPED_STORAGE:
528 t->num_mapped_columns++;
529 col->index_offset = t->index_entry_size;
530 t->index_entry_size += 8;
531 break;
532 case OSL_NO_STORAGE:
533 col->volatile_num = t->num_volatile_columns;
534 t->num_volatile_columns++;
535 break;
536 }
537 if (sf & OSL_RBTREE) {
538 col->rbtree_num = t->num_rbtrees;
539 t->num_rbtrees++;
540 if ((sf & OSL_UNIQUE) && (st == OSL_MAPPED_STORAGE)) {
541 if (!have_disk_storage_name_column)
542 t->disk_storage_name_column = i;
543 have_disk_storage_name_column = 1;
544 }
545 }
546 }
547 ret = -E_NO_UNIQUE_RBTREE_COLUMN;
548 if (t->num_disk_storage_columns && !have_disk_storage_name_column)
549 goto err;
550 ret = -E_NO_RBTREE_COL;
551 if (!t->num_rbtrees)
552 goto err;
553 /* success */
554 PARA_DEBUG_LOG("OK. Index entry size: %u\n", t->index_entry_size);
555 ret = init_column_descriptions(t);
556 if (ret < 0)
557 goto err;
558 *table_ptr = t;
559 return 1;
560 err:
561 free(t->columns);
562 free(t);
563 return ret;
564 }
565
566 /**
567 * Read the table description from index header.
568 *
569 * \param map The memory mapping of the index file.
570 * \param desc The values found in the index header are returned here.
571 *
572 * Read the index header, check for the paraslash magic string and the table version number.
573 * Read all information stored in the index header into \a desc.
574 *
575 * \return Positive on success, negative on errors.
576 *
577 * \sa struct osl_table_description, osl_create_table.
578 */
579 int read_table_desc(struct osl_object *map, struct osl_table_description *desc)
580 {
581 char *buf = map->data;
582 uint8_t version;
583 uint16_t header_size;
584 int ret, i;
585 unsigned offset;
586 struct osl_column_description *cd;
587
588 if (map->size < MIN_INDEX_HEADER_SIZE(1))
589 return -E_SHORT_TABLE;
590 if (strncmp(buf + IDX_PARA_MAGIC, PARA_MAGIC, strlen(PARA_MAGIC)))
591 return -E_NO_MAGIC;
592 version = read_u8(buf + IDX_VERSION);
593 if (version < MIN_TABLE_VERSION || version > MAX_TABLE_VERSION)
594 return -E_VERSION_MISMATCH;
595 desc->num_columns = read_u8(buf + IDX_TABLE_FLAGS);
596 desc->flags = read_u8(buf + IDX_TABLE_FLAGS);
597 desc->num_columns = read_u16(buf + IDX_NUM_COLUMNS);
598 PARA_DEBUG_LOG("%u columns\n", desc->num_columns);
599 if (!desc->num_columns)
600 return -E_NO_COLUMNS;
601 header_size = read_u16(buf + IDX_HEADER_SIZE);
602 if (map->size < header_size)
603 return -E_BAD_SIZE;
604 desc->column_descriptions = para_calloc(desc->num_columns
605 * sizeof(struct osl_column_description));
606 offset = IDX_COLUMN_DESCRIPTIONS;
607 FOR_EACH_COLUMN(i, desc, cd) {
608 char *null_byte;
609
610 ret = -E_SHORT_TABLE;
611 if (map->size < offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE) {
612 PARA_ERROR_LOG("map size = %zu < %u = offset + min desc size\n",
613 map->size, offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE);
614 goto err;
615 }
616 cd->storage_type = read_u16(buf + offset + IDX_CD_STORAGE_TYPE);
617 cd->storage_flags = read_u16(buf + offset +
618 IDX_CD_STORAGE_FLAGS);
619 cd->data_size = read_u32(buf + offset + IDX_CD_DATA_SIZE);
620 null_byte = memchr(buf + offset + IDX_CD_NAME, '\0',
621 map->size - offset - IDX_CD_NAME);
622 ret = -E_INDEX_CORRUPTION;
623 if (!null_byte)
624 goto err;
625 cd->name = para_strdup(buf + offset + IDX_CD_NAME);
626 offset += index_column_description_size(cd->name);
627 }
628 if (offset != header_size) {
629 ret = -E_INDEX_CORRUPTION;
630 PARA_ERROR_LOG("real header size = %u != %u = stored header size\n",
631 offset, header_size);
632 goto err;
633 }
634 return 1;
635 err:
636 FOR_EACH_COLUMN(i, desc, cd)
637 free(cd->name);
638 return ret;
639 }
640
641 /*
642 * check whether the table description given by \p t->desc matches the on-disk
643 * table structure stored in the index of \a t.
644 */
645 static int compare_table_descriptions(struct osl_table *t)
646 {
647 int i, ret;
648 struct osl_table_description desc;
649 const struct osl_column_description *cd1, *cd2;
650
651 /* read the on-disk structure into desc */
652 ret = read_table_desc(&t->index_map, &desc);
653 if (ret < 0)
654 return ret;
655 ret = -E_BAD_TABLE_FLAGS;
656 if (desc.flags != t->desc->flags)
657 goto out;
658 ret = E_BAD_COLUMN_NUM;
659 if (desc.num_columns != t->desc->num_columns)
660 goto out;
661 FOR_EACH_COLUMN(i, t->desc, cd1) {
662 cd2 = get_column_description(&desc, i);
663 ret = -E_BAD_STORAGE_TYPE;
664 if (cd1->storage_type != cd2->storage_type)
665 goto out;
666 ret = -E_BAD_STORAGE_FLAGS;
667 if (cd1->storage_flags != cd2->storage_flags) {
668 PARA_ERROR_LOG("sf1 = %u != %u = sf2\n",
669 cd1->storage_flags, cd2->storage_flags);
670 goto out;
671 }
672 ret = -E_BAD_DATA_SIZE;
673 if (cd1->storage_flags & OSL_FIXED_SIZE)
674 if (cd1->data_size != cd2->data_size)
675 goto out;
676 ret = -E_BAD_COLUMN_NAME;
677 if (strcmp(cd1->name, cd2->name))
678 goto out;
679 }
680 PARA_DEBUG_LOG("table description of '%s' matches on-disk data, good\n",
681 t->desc->name);
682 ret = 1;
683 out:
684 FOR_EACH_COLUMN(i, &desc, cd1)
685 free(cd1->name);
686 free(desc.column_descriptions);
687 return ret;
688 }
689
690 static int create_table_index(struct osl_table *t)
691 {
692 char *buf, *filename;
693 int i, ret;
694 size_t size = t->index_header_size;
695 const struct osl_column_description *cd;
696 unsigned offset;
697
698 PARA_INFO_LOG("creating %zu byte index for table %s\n", size,
699 t->desc->name);
700 buf = para_calloc(size);
701 sprintf(buf + IDX_PARA_MAGIC, "%s", PARA_MAGIC);
702 write_u8(buf + IDX_TABLE_FLAGS, t->desc->flags);
703 write_u8(buf + IDX_DIRTY_FLAG, 0);
704 write_u8(buf + IDX_VERSION, CURRENT_TABLE_VERSION);
705 write_u16(buf + IDX_NUM_COLUMNS, t->desc->num_columns);
706 write_u16(buf + IDX_HEADER_SIZE, t->index_header_size);
707 offset = IDX_COLUMN_DESCRIPTIONS;
708 FOR_EACH_COLUMN(i, t->desc, cd) {
709 write_u16(buf + offset + IDX_CD_STORAGE_TYPE,
710 cd->storage_type);
711 write_u16(buf + offset + IDX_CD_STORAGE_FLAGS,
712 cd->storage_flags);
713 if (cd->storage_flags & OSL_FIXED_SIZE)
714 write_u32(buf + offset + IDX_CD_DATA_SIZE,
715 cd->data_size);
716 strcpy(buf + offset + IDX_CD_NAME, cd->name);
717 offset += index_column_description_size(cd->name);
718 }
719 assert(offset = size);
720 filename = index_filename(t->desc);
721 ret = para_write_file(filename, buf, size);
722 free(buf);
723 free(filename);
724 return ret;
725 }
726
727 /**
728 * Create a new osl table.
729 *
730 * \param desc Pointer to the table description.
731 *
732 * \return Positive on success, negative on errors. Possible errors include: \p
733 * E_BAD_TABLE_DESC, \p E_BAD_DB_DIR, \p E_BAD_BASENAME, \p E_NO_COMPARE_FUNC, \p
734 * E_NO_COLUMN_NAME, \p E_DUPLICATE_COL_NAME, \p E_MKDIR, any errors returned
735 * by para_open().
736 */
737 int osl_create_table(const struct osl_table_description *desc)
738 {
739 const struct osl_column_description *cd;
740 char *table_dir = NULL, *filename;
741 struct osl_table *t;
742 int i, ret = init_table_structure(desc, &t);
743
744 if (ret < 0)
745 return ret;
746 PARA_INFO_LOG("creating %s\n", desc->name);
747 FOR_EACH_COLUMN(i, t->desc, cd) {
748 if (cd->storage_type == OSL_NO_STORAGE)
749 continue;
750 if (!table_dir) {
751 ret = para_mkdir(desc->dir, 0777);
752 if (ret < 0 && ret != -E_EXIST)
753 goto out;
754 table_dir = make_message("%s/%s", desc->dir,
755 desc->name);
756 ret = para_mkdir(table_dir, 0777);
757 if (ret < 0)
758 goto out;
759 }
760 filename = column_filename(t, i);
761 PARA_INFO_LOG("filename: %s\n", filename);
762 if (cd->storage_type == OSL_MAPPED_STORAGE) {
763 ret = para_open(filename, O_RDWR | O_CREAT | O_EXCL,
764 0644);
765 free(filename);
766 if (ret < 0)
767 goto out;
768 close(ret);
769 continue;
770 }
771 /* DISK STORAGE */
772 ret = para_mkdir(filename, 0777);
773 free(filename);
774 if (ret < 0)
775 goto out;
776 }
777 if (t->num_mapped_columns) {
778 ret = create_table_index(t);
779 if (ret < 0)
780 goto out;
781 }
782 ret = 1;
783 out:
784 free(table_dir);
785 free(t->columns);
786 free(t);
787 return ret;
788 }
789
790 static int table_is_dirty(struct osl_table *t)
791 {
792 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
793 uint8_t dirty = read_u8(buf) & 0x1;
794 return !!dirty;
795 }
796
797 static void mark_table_dirty(struct osl_table *t)
798 {
799 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
800 write_u8(buf, read_u8(buf) | 1);
801 }
802
803 static void mark_table_clean(struct osl_table *t)
804 {
805 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
806 write_u8(buf, read_u8(buf) & 0xfe);
807 }
808
809 /**
810 * Unmap all mapped files of an osl table.
811 *
812 * \param t Pointer to a mapped table.
813 * \param flags Options for unmapping.
814 *
815 * \return Positive on success, negative on errors. Possible errors include:
816 * E_NOT_MAPPED, E_MUNMAP.
817 *
818 * \sa map_table(), enum osl_close_flags, para_munmap().
819 */
820 int unmap_table(struct osl_table *t, enum osl_close_flags flags)
821 {
822 unsigned i;
823 const struct osl_column_description *cd;
824 int ret;
825
826 if (!t->num_mapped_columns) /* can this ever happen? */
827 return 1;
828 PARA_DEBUG_LOG("unmapping table '%s'\n", t->desc->name);
829 if (!t->index_map.data)
830 return -E_NOT_MAPPED;
831 if (flags & OSL_MARK_CLEAN)
832 mark_table_clean(t);
833 ret = para_munmap(t->index_map.data, t->index_map.size);
834 if (ret < 0)
835 return ret;
836 t->index_map.data = NULL;
837 if (!t->num_rows)
838 return 1;
839 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
840 struct osl_object map = t->columns[i].data_map;
841 if (!map.data)
842 continue;
843 ret = para_munmap(map.data, map.size);
844 if (ret < 0)
845 return ret;
846 map.data = NULL;
847 }
848 return 1;
849 }
850
851 /**
852 * Map the index file and all columns of type \p OSL_MAPPED_STORAGE into memory.
853 *
854 * \param t Pointer to an initialized table structure.
855 * \param flags Mapping options.
856 *
857 * \return Negative return value on errors; on success the number of rows
858 * (including invalid rows) is returned.
859 *
860 * \sa unmap_table(), enum map_table_flags, osl_open_table(), mmap(2).
861 */
862 int map_table(struct osl_table *t, enum map_table_flags flags)
863 {
864 char *filename;
865 const struct osl_column_description *cd;
866 int i = 0, ret, num_rows = 0;
867
868 if (!t->num_mapped_columns)
869 return 0;
870 if (t->index_map.data)
871 return -E_ALREADY_MAPPED;
872 filename = index_filename(t->desc);
873 PARA_DEBUG_LOG("mapping table '%s' (index: %s)\n", t->desc->name, filename);
874 ret = mmap_full_file(filename, flags & MAP_TBL_FL_MAP_RDONLY?
875 O_RDONLY : O_RDWR, &t->index_map);
876 free(filename);
877 if (ret < 0)
878 return ret;
879 if (flags & MAP_TBL_FL_VERIFY_INDEX) {
880 ret = compare_table_descriptions(t);
881 if (ret < 0)
882 goto err;
883 }
884 ret = -E_BUSY;
885 if (!(flags & MAP_TBL_FL_IGNORE_DIRTY)) {
886 if (table_is_dirty(t)) {
887 PARA_ERROR_LOG("%s is dirty\n", t->desc->name);
888 goto err;
889 }
890 }
891 mark_table_dirty(t);
892 num_rows = table_num_rows(t);
893 if (!num_rows)
894 return num_rows;
895 /* map data files */
896 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
897 struct stat statbuf;
898 filename = column_filename(t, i);
899 ret = -E_STAT;
900 if (stat(filename, &statbuf) < 0) {
901 free(filename);
902 goto err;
903 }
904 if (!(S_IFREG & statbuf.st_mode)) {
905 free(filename);
906 goto err;
907 }
908 ret = mmap_full_file(filename, O_RDWR,
909 &t->columns[i].data_map);
910 free(filename);
911 if (ret < 0)
912 goto err;
913 }
914 return num_rows;
915 err: /* unmap what is already mapped */
916 for (i--; i >= 0; i--) {
917 struct osl_object map = t->columns[i].data_map;
918 para_munmap(map.data, map.size);
919 map.data = NULL;
920 }
921 para_munmap(t->index_map.data, t->index_map.size);
922 t->index_map.data = NULL;
923 return ret;
924 }
925
926 /**
927 * Retrieve a mapped object by row and column number.
928 *
929 * \param t Pointer to an open osl table.
930 * \param col_num Number of the mapped column containing the object to retrieve.
931 * \param row_num Number of the row containing the object to retrieve.
932 * \param obj The result is returned here.
933 *
934 * It is considered an error if \a col_num does not refer to a column
935 * of storage type \p OSL_MAPPED_STORAGE.
936 *
937 * \return Positive on success, negative on errors. Possible errors include:
938 * \p E_BAD_ID, \p E_INVALID_OBJECT.
939 *
940 * \sa osl_storage_type.
941 */
942 int get_mapped_object(const struct osl_table *t, unsigned col_num,
943 uint32_t row_num, struct osl_object *obj)
944 {
945 struct osl_column *col = &t->columns[col_num];
946 uint32_t offset;
947 char *header;
948 char *index_entry;
949 int ret;
950
951 if (t->num_rows <= row_num)
952 return -E_BAD_ID;
953 ret = get_index_entry(t, row_num, col_num, &index_entry);
954 if (ret < 0)
955 return ret;
956 offset = read_u32(index_entry);
957 obj->size = read_u32(index_entry + 4) - 1;
958 header = col->data_map.data + offset;
959 obj->data = header + 1;
960 if (read_u8(header) == 0xff) {
961 PARA_ERROR_LOG("col %u, size %zu, offset %u\n", col_num,
962 obj->size, offset);
963 return -E_INVALID_OBJECT;
964 }
965 return 1;
966 }
967
968 static int search_rbtree(const struct osl_object *obj,
969 const struct osl_table *t, unsigned col_num,
970 struct rb_node **result, struct rb_node ***rb_link)
971 {
972 struct osl_column *col = &t->columns[col_num];
973 struct rb_node **new = &col->rbtree.rb_node, *parent = NULL;
974 const struct osl_column_description *cd =
975 get_column_description(t->desc, col_num);
976 enum osl_storage_type st = cd->storage_type;
977 while (*new) {
978 struct osl_row *this_row = get_row_pointer(*new,
979 col->rbtree_num);
980 int ret;
981 struct osl_object this_obj;
982 parent = *new;
983 if (st == OSL_MAPPED_STORAGE) {
984 ret = get_mapped_object(t, col_num, this_row->id,
985 &this_obj);
986 if (ret < 0)
987 return ret;
988 } else
989 this_obj = this_row->volatile_objects[col->volatile_num];
990 ret = cd->compare_function(obj, &this_obj);
991 if (!ret) {
992 if (result)
993 *result = get_rb_node_pointer(this_row,
994 col->rbtree_num);
995 return 1;
996 }
997 if (ret < 0)
998 new = &((*new)->rb_left);
999 else
1000 new = &((*new)->rb_right);
1001 }
1002 if (result)
1003 *result = parent;
1004 if (rb_link)
1005 *rb_link = new;
1006 return -E_RB_KEY_NOT_FOUND;
1007 }
1008
1009 static int insert_rbtree(struct osl_table *t, unsigned col_num,
1010 const struct osl_row *row, const struct osl_object *obj)
1011 {
1012 struct rb_node *parent, **rb_link;
1013 unsigned rbtree_num;
1014 struct rb_node *n;
1015 int ret = search_rbtree(obj, t, col_num, &parent, &rb_link);
1016
1017 if (ret > 0)
1018 return -E_RB_KEY_EXISTS;
1019 rbtree_num = t->columns[col_num].rbtree_num;
1020 n = get_rb_node_pointer(row, rbtree_num);
1021 rb_link_node(n, parent, rb_link);
1022 rb_insert_color(n, &t->columns[col_num].rbtree);
1023 return 1;
1024 }
1025
1026 static void remove_rb_node(struct osl_table *t, unsigned col_num,
1027 const struct osl_row *row)
1028 {
1029 struct osl_column *col = &t->columns[col_num];
1030 const struct osl_column_description *cd =
1031 get_column_description(t->desc, col_num);
1032 enum osl_storage_flags sf = cd->storage_flags;
1033 struct rb_node *victim, *splice_out_node, *tmp;
1034 if (!(sf & OSL_RBTREE))
1035 return;
1036 /*
1037 * Which node is removed/spliced out actually depends on how many
1038 * children the victim node has: If it has no children, it gets
1039 * deleted. If it has one child, it gets spliced out. If it has two
1040 * children, its successor (which has at most a right child) gets
1041 * spliced out.
1042 */
1043 victim = get_rb_node_pointer(row, col->rbtree_num);
1044 if (victim->rb_left && victim->rb_right)
1045 splice_out_node = rb_next(victim);
1046 else
1047 splice_out_node = victim;
1048 /* Go up to the root and decrement the size of each node in the path. */
1049 for (tmp = splice_out_node; tmp; tmp = rb_parent(tmp))
1050 tmp->size--;
1051 rb_erase(victim, &col->rbtree);
1052 }
1053
1054 static int add_row_to_rbtrees(struct osl_table *t, uint32_t id,
1055 struct osl_object *volatile_objs, struct osl_row **row_ptr)
1056 {
1057 unsigned i;
1058 int ret;
1059 struct osl_row *row = allocate_row(t->num_rbtrees);
1060 const struct osl_column_description *cd;
1061
1062 row->id = id;
1063 row->volatile_objects = volatile_objs;
1064 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
1065 if (cd->storage_type == OSL_MAPPED_STORAGE) {
1066 struct osl_object obj;
1067 ret = get_mapped_object(t, i, id, &obj);
1068 if (ret < 0)
1069 goto err;
1070 ret = insert_rbtree(t, i, row, &obj);
1071 } else { /* volatile */
1072 const struct osl_object *obj
1073 = volatile_objs + t->columns[i].volatile_num;
1074 ret = insert_rbtree(t, i, row, obj);
1075 }
1076 if (ret < 0)
1077 goto err;
1078 }
1079 if (row_ptr)
1080 *row_ptr = row;
1081 return 1;
1082 err: /* rollback changes, i.e. remove added entries from rbtrees */
1083 while (i)
1084 remove_rb_node(t, i--, row);
1085 free(row);
1086 return ret;
1087 }
1088
1089 static void free_volatile_objects(const struct osl_table *t,
1090 enum osl_close_flags flags)
1091 {
1092 int i, j;
1093 struct rb_node *n;
1094 struct osl_column *rb_col;
1095 const struct osl_column_description *cd;
1096
1097 if (!t->num_volatile_columns)
1098 return;
1099 /* find the first rbtree column (any will do) */
1100 FOR_EACH_RBTREE_COLUMN(i, t, cd)
1101 break;
1102 rb_col = t->columns + i;
1103 /* walk that rbtree and free all volatile objects */
1104 for (n = rb_first(&rb_col->rbtree); n; n = rb_next(n)) {
1105 struct osl_row *r = get_row_pointer(n, rb_col->rbtree_num);
1106 if (flags & OSL_FREE_VOLATILE)
1107 for (j = 0; j < t->num_volatile_columns; j++)
1108 free(r->volatile_objects[j].data);
1109 free(r->volatile_objects);
1110 }
1111 }
1112
1113 /**
1114 * Erase all rbtree nodes and free resources.
1115 *
1116 * \param t Pointer to an open osl table.
1117 *
1118 * This function is called by osl_close_table().
1119 */
1120 void clear_rbtrees(struct osl_table *t)
1121 {
1122 const struct osl_column_description *cd;
1123 unsigned i, rbtrees_cleared = 0;
1124
1125 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
1126 struct osl_column *col = &t->columns[i];
1127 struct rb_node *n;
1128 rbtrees_cleared++;
1129 for (n = rb_first(&col->rbtree); n;) {
1130 struct osl_row *r;
1131 rb_erase(n, &col->rbtree);
1132 if (rbtrees_cleared == t->num_rbtrees) {
1133 r = get_row_pointer(n, col->rbtree_num);
1134 n = rb_next(n);
1135 free(r);
1136 } else
1137 n = rb_next(n);
1138 }
1139 }
1140
1141 }
1142
1143 /**
1144 * Close an osl table.
1145 *
1146 * \param t Pointer to the table to be closed.
1147 * \param flags Options for what should be cleaned up.
1148 *
1149 * If osl_open_table() succeeds, the resulting table pointer must later be
1150 * passed to this function in order to flush all changes to the filesystem and
1151 * to free the resources that were allocated by osl_open_table().
1152 *
1153 * \return Positive on success, negative on errors. Possible errors: \p E_BAD_TABLE,
1154 * errors returned by unmap_table().
1155 *
1156 * \sa osl_open_table(), unmap_table().
1157 */
1158 int osl_close_table(struct osl_table *t, enum osl_close_flags flags)
1159 {
1160 int ret;
1161
1162 if (!t)
1163 return -E_BAD_TABLE;
1164 free_volatile_objects(t, flags);
1165 clear_rbtrees(t);
1166 ret = unmap_table(t, flags);
1167 if (ret < 0)
1168 PARA_ERROR_LOG("unmap_table failed: %d\n", ret);
1169 free(t->columns);
1170 free(t);
1171 return ret;
1172 }
1173
1174 /**
1175 * Find out whether the given row number corresponds to an invalid row.
1176 *
1177 * \param t Pointer to the osl table.
1178 * \param row_num The number of the row in question.
1179 *
1180 * By definition, a row is considered invalid if all its index entries
1181 * are invalid.
1182 *
1183 * \return Positive if \a row_num corresponds to an invalid row,
1184 * zero if it corresponds to a valid row, negative on errors.
1185 */
1186 int row_is_invalid(struct osl_table *t, uint32_t row_num)
1187 {
1188 char *index_entry;
1189 int i, ret = get_index_entry_start(t, row_num, &index_entry);
1190
1191 if (ret < 0)
1192 return ret;
1193 for (i = 0; i < t->index_entry_size; i++) {
1194 if ((unsigned char)index_entry[i] != 0xff)
1195 return 0;
1196 }
1197 PARA_INFO_LOG("row %d is invalid\n", row_num);
1198 return 1;
1199 }
1200
1201 /**
1202 * Invalidate a row of an osl table.
1203 *
1204 * \param t Pointer to an open osl table.
1205 * \param row_num Number of the row to mark as invalid.
1206 *
1207 * This function marks each mapped object in the index entry of \a row as
1208 * invalid.
1209 *
1210 * \return Positive on success, negative on errors.
1211 */
1212 int mark_row_invalid(struct osl_table *t, uint32_t row_num)
1213 {
1214 char *index_entry;
1215 int i, ret = get_index_entry_start(t, row_num, &index_entry);
1216
1217 PARA_INFO_LOG("marking row %d as invalid\n", row_num);
1218 if (ret < 0)
1219 return ret;
1220 for (i = 0; i < t->index_entry_size; i++)
1221 index_entry[i] = 0xff;
1222 return 1;
1223 }
1224
1225 /**
1226 * Initialize all rbtrees and compute number of invalid rows.
1227 *
1228 * \param t The table containing the rbtrees to be initialized.
1229 *
1230 * \return Positive on success, negative on errors.
1231 */
1232 int init_rbtrees(struct osl_table *t)
1233 {
1234 int i, ret;
1235 const struct osl_column_description *cd;
1236
1237 /* create rbtrees */
1238 FOR_EACH_RBTREE_COLUMN(i, t, cd)
1239 t->columns[i].rbtree = RB_ROOT;
1240 /* add valid rows to rbtrees */
1241 t->num_invalid_rows = 0;
1242 for (i = 0; i < t->num_rows; i++) {
1243 ret = row_is_invalid(t, i);
1244 if (ret < 0)
1245 return ret;
1246 if (ret) {
1247 t->num_invalid_rows++;
1248 continue;
1249 }
1250 ret = add_row_to_rbtrees(t, i, NULL, NULL);
1251 if (ret < 0)
1252 return ret;
1253 }
1254 return 1;
1255 }
1256
1257 /**
1258 * Open an osl table.
1259 *
1260 * Each osl table must be opened before its data can be accessed.
1261 *
1262 * \param table_desc Describes the table to be opened.
1263 * \param result Contains a pointer to the open table on success.
1264 *
1265 * The table description given by \a desc should coincide with the
1266 * description used at creation time.
1267 *
1268 * \return Positive on success, negative on errors. Possible errors include:
1269 * errors returned by init_table_structure(), \p E_NOENT, \p E_STAT, \p \p
1270 * E_NOTDIR, \p E_BAD_TABLE_DESC, \p E_BAD_DB_DIR, \p E_NO_COMPARE_FUNC, \p
1271 * E_NO_COLUMN_NAME, errors returned by init_rbtrees().
1272 */
1273 int osl_open_table(const struct osl_table_description *table_desc,
1274 struct osl_table **result)
1275 {
1276 int i, ret;
1277 struct osl_table *t;
1278 const struct osl_column_description *cd;
1279
1280 PARA_INFO_LOG("opening table %s\n", table_desc->name);
1281 ret = init_table_structure(table_desc, &t);
1282 if (ret < 0)
1283 return ret;
1284 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1285 /* check if directory exists */
1286 char *dirname = column_filename(t, i);
1287 struct stat statbuf;
1288 ret = stat(dirname, &statbuf);
1289 free(dirname);
1290 if (ret < 0) {
1291 if (errno == ENOENT)
1292 ret = -E_NOENT;
1293 else
1294 ret = -E_STAT;
1295 goto err;
1296 }
1297 ret = -E_NOTDIR;
1298 if (!S_ISDIR(statbuf.st_mode))
1299 goto err;
1300 }
1301 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1302 if (ret < 0)
1303 goto err;
1304 t->num_rows = ret;
1305 PARA_DEBUG_LOG("num rows: %d\n", t->num_rows);
1306 ret = init_rbtrees(t);
1307 if (ret < 0) {
1308 osl_close_table(t, OSL_MARK_CLEAN); /* ignore further errors */
1309 return ret;
1310 }
1311 *result = t;
1312 return 1;
1313 err:
1314 free(t->columns);
1315 free(t);
1316 return ret;
1317 }
1318
1319 static int create_disk_storage_object_dir(const struct osl_table *t,
1320 unsigned col_num, const char *ds_name)
1321 {
1322 char *dirname;
1323 int ret;
1324
1325 if (!(t->desc->flags & OSL_LARGE_TABLE))
1326 return 1;
1327 dirname = disk_storage_dirname(t, col_num, ds_name);
1328 ret = para_mkdir(dirname, 0777);
1329 free(dirname);
1330 if (ret < 0 && ret != -E_EXIST)
1331 return ret;
1332 return 1;
1333 }
1334
1335 static int write_disk_storage_file(const struct osl_table *t, unsigned col_num,
1336 const struct osl_object *obj, const char *ds_name)
1337 {
1338 int ret;
1339 char *filename;
1340
1341 ret = create_disk_storage_object_dir(t, col_num, ds_name);
1342 if (ret < 0)
1343 return ret;
1344 filename = disk_storage_path(t, col_num, ds_name);
1345 ret = para_write_file(filename, obj->data, obj->size);
1346 free(filename);
1347 return ret;
1348 }
1349
1350 static int append_map_file(const struct osl_table *t, unsigned col_num,
1351 const struct osl_object *obj, uint32_t *new_size)
1352 {
1353 char *filename = column_filename(t, col_num);
1354 int ret;
1355 char header = 0; /* zero means valid object */
1356
1357 // PARA_DEBUG_LOG("appending %zu + 1 byte\n", obj->size);
1358 ret = append_file(filename, &header, 1, obj->data, obj->size,
1359 new_size);
1360 free(filename);
1361 return ret;
1362 }
1363
1364 static int append_index_entry(const struct osl_table *t, char *new_index_entry)
1365 {
1366 char *filename;
1367 int ret;
1368
1369 if (!t->num_mapped_columns)
1370 return 1;
1371 filename = index_filename(t->desc);
1372 // PARA_DEBUG_LOG("appending %u bytes\n", t->index_entry_size);
1373 ret = append_file(filename, NULL, 0, new_index_entry,
1374 t->index_entry_size, NULL);
1375 free(filename);
1376 return ret;
1377 }
1378
1379 /**
1380 * A wrapper for truncate(2)
1381 *
1382 * \param path Name of the regular file to truncate
1383 * \param size Number of bytes to \b shave \b off
1384 *
1385 * Truncate the regular file named by \a path by \a size bytes.
1386 *
1387 * \return Positive on success, negative on errors. Possible errors include: \p
1388 * E_STAT, \p E_BAD_SIZE, \p E_TRUNC.
1389 *
1390 * \sa truncate(2)
1391 */
1392 int para_truncate(const char *path, off_t size)
1393 {
1394 int ret;
1395 struct stat statbuf;
1396
1397 ret = -E_STAT;
1398 if (stat(path, &statbuf) < 0)
1399 goto out;
1400 ret = -E_BAD_SIZE;
1401 if (statbuf.st_size < size)
1402 goto out;
1403 ret = -E_TRUNC;
1404 if (truncate(path, statbuf.st_size - size) < 0)
1405 goto out;
1406 ret = 1;
1407 out:
1408 return ret;
1409 }
1410
1411 static int truncate_mapped_file(const struct osl_table *t, unsigned col_num,
1412 off_t size)
1413 {
1414 char *filename = column_filename(t, col_num);
1415 int ret = para_truncate(filename, size);
1416 free(filename);
1417 return ret;
1418 }
1419
1420 static int delete_disk_storage_file(const struct osl_table *t, unsigned col_num,
1421 const char *ds_name)
1422 {
1423 char *dirname, *filename = disk_storage_path(t, col_num, ds_name);
1424 int ret = unlink(filename);
1425
1426 PARA_INFO_LOG("deleted %s\n", filename);
1427 free(filename);
1428 if (ret < 0) {
1429 if (errno == ENOENT)
1430 return -E_NOENT;
1431 return -E_UNLINK;
1432 }
1433 if (!(t->desc->flags & OSL_LARGE_TABLE))
1434 return 1;
1435 dirname = disk_storage_dirname(t, col_num, ds_name);
1436 rmdir(dirname);
1437 free(dirname);
1438 return 1;
1439 }
1440
1441 /**
1442 * Add a new row to an osl table and retrieve this row.
1443 *
1444 * \param t Pointer to an open osl table.
1445 * \param objects Array of objects to be added.
1446 * \param row Result pointer.
1447 *
1448 * The \a objects parameter must point to an array containing one object per
1449 * column. The order of the objects in the array is given by the table
1450 * description of \a table. Several sanity checks are performed during object
1451 * insertion and the function returns without modifying the table if any of
1452 * these tests fail. In fact, it is atomic in the sense that it either
1453 * succeeds or leaves the table unchanged (i.e. either all or none of the
1454 * objects are added to the table).
1455 *
1456 * It is considered an error if an object is added to a column with associated
1457 * rbtree if this object is equal to an object already contained in that column
1458 * (i.e. the compare function for the column's rbtree returns zero).
1459 *
1460 * Possible errors include: \p E_RB_KEY_EXISTS, \p E_BAD_DATA_SIZE.
1461 *
1462 * \return Positive on success, negative on errors.
1463 *
1464 * \sa struct osl_table_description, osl_compare_func, osl_add_row().
1465 */
1466 int osl_add_and_get_row(struct osl_table *t, struct osl_object *objects,
1467 struct osl_row **row)
1468 {
1469 int i, ret;
1470 char *ds_name = NULL;
1471 struct rb_node **rb_parents = NULL, ***rb_links = NULL;
1472 char *new_index_entry = NULL;
1473 struct osl_object *volatile_objs = NULL;
1474 const struct osl_column_description *cd;
1475
1476 if (!t)
1477 return -E_BAD_TABLE;
1478 rb_parents = para_malloc(t->num_rbtrees * sizeof(struct rn_node*));
1479 rb_links = para_malloc(t->num_rbtrees * sizeof(struct rn_node**));
1480 if (t->num_mapped_columns)
1481 new_index_entry = para_malloc(t->index_entry_size);
1482 /* pass 1: sanity checks */
1483 // PARA_DEBUG_LOG("sanity tests: %p:%p\n", objects[0].data,
1484 // objects[1].data);
1485 FOR_EACH_COLUMN(i, t->desc, cd) {
1486 enum osl_storage_type st = cd->storage_type;
1487 enum osl_storage_flags sf = cd->storage_flags;
1488
1489 // ret = -E_NULL_OBJECT;
1490 // if (!objects[i])
1491 // goto out;
1492 if (st == OSL_DISK_STORAGE)
1493 continue;
1494 if (sf & OSL_RBTREE) {
1495 unsigned rbtree_num = t->columns[i].rbtree_num;
1496 ret = -E_RB_KEY_EXISTS;
1497 // PARA_DEBUG_LOG("checking whether %p exists\n",
1498 // objects[i].data);
1499 if (search_rbtree(objects + i, t, i,
1500 &rb_parents[rbtree_num],
1501 &rb_links[rbtree_num]) > 0)
1502 goto out;
1503 }
1504 if (sf & OSL_FIXED_SIZE) {
1505 // PARA_DEBUG_LOG("fixed size. need: %zu, have: %d\n",
1506 // objects[i].size, cd->data_size);
1507 ret = -E_BAD_DATA_SIZE;
1508 if (objects[i].size != cd->data_size)
1509 goto out;
1510 }
1511 }
1512 if (t->num_disk_storage_columns)
1513 ds_name = disk_storage_name_of_object(t,
1514 &objects[t->disk_storage_name_column]);
1515 ret = unmap_table(t, OSL_MARK_CLEAN);
1516 if (ret < 0)
1517 goto out;
1518 // PARA_DEBUG_LOG("sanity tests passed%s\n", "");
1519 /* pass 2: create data files, append map data */
1520 FOR_EACH_COLUMN(i, t->desc, cd) {
1521 enum osl_storage_type st = cd->storage_type;
1522 if (st == OSL_NO_STORAGE)
1523 continue;
1524 if (st == OSL_MAPPED_STORAGE) {
1525 uint32_t new_size;
1526 struct osl_column *col = &t->columns[i];
1527 // PARA_DEBUG_LOG("appending object of size %zu\n",
1528 // objects[i].size);
1529 ret = append_map_file(t, i, objects + i, &new_size);
1530 if (ret < 0)
1531 goto rollback;
1532 update_index_entry(new_index_entry, col, new_size,
1533 objects[i].size);
1534 continue;
1535 }
1536 /* DISK_STORAGE */
1537 ret = write_disk_storage_file(t, i, objects + i, ds_name);
1538 if (ret < 0)
1539 goto rollback;
1540 }
1541 ret = append_index_entry(t, new_index_entry);
1542 if (ret < 0)
1543 goto rollback;
1544 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1545 if (ret < 0) { /* truncate index and rollback changes */
1546 char *filename = index_filename(t->desc);
1547 para_truncate(filename, t->index_entry_size);
1548 free(filename);
1549 goto rollback;
1550 }
1551 /* pass 3: add entry to rbtrees */
1552 if (t->num_volatile_columns) {
1553 volatile_objs = para_calloc(t->num_volatile_columns
1554 * sizeof(struct osl_object));
1555 FOR_EACH_VOLATILE_COLUMN(i, t, cd)
1556 volatile_objs[t->columns[i].volatile_num] = objects[i];
1557 }
1558 t->num_rows++;
1559 // PARA_DEBUG_LOG("adding new entry as row #%d\n", t->num_rows - 1);
1560 ret = add_row_to_rbtrees(t, t->num_rows - 1, volatile_objs, row);
1561 if (ret < 0)
1562 goto out;
1563 // PARA_DEBUG_LOG("added new entry as row #%d\n", t->num_rows - 1);
1564 ret = 1;
1565 goto out;
1566 rollback: /* rollback all changes made, ignore further errors */
1567 for (i--; i >= 0; i--) {
1568 cd = get_column_description(t->desc, i);
1569 enum osl_storage_type st = cd->storage_type;
1570 if (st == OSL_NO_STORAGE)
1571 continue;
1572
1573 if (st == OSL_MAPPED_STORAGE)
1574 truncate_mapped_file(t, i, objects[i].size);
1575 else /* disk storage */
1576 delete_disk_storage_file(t, i, ds_name);
1577 }
1578 /* ignore error and return previous error value */
1579 map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1580 out:
1581 free(new_index_entry);
1582 free(ds_name);
1583 free(rb_parents);
1584 free(rb_links);
1585 return ret;
1586 }
1587
1588 /**
1589 * Add a new row to an osl table.
1590 *
1591 * \param t Same meaning as osl_add_and_get_row().
1592 * \param objects Same meaning as osl_add_and_get_row().
1593 *
1594 * \return The return value of the underlying call to osl_add_and_get_row().
1595 *
1596 * This is equivalent to osl_add_and_get_row(t, objects, NULL).
1597 */
1598 int osl_add_row(struct osl_table *t, struct osl_object *objects)
1599 {
1600 return osl_add_and_get_row(t, objects, NULL);
1601 }
1602
1603 /**
1604 * Retrieve an object identified by row and column
1605 *
1606 * \param t Pointer to an open osl table.
1607 * \param r Pointer to the row.
1608 * \param col_num The column number.
1609 * \param object The result pointer.
1610 *
1611 * The column determined by \a col_num must be of type \p OSL_MAPPED_STORAGE
1612 * or \p OSL_NO_STORAGE, i.e. no disk storage objects may be retrieved by this
1613 * function.
1614 *
1615 * \return Positive if object was found, negative on errors. Possible errors
1616 * include: \p E_BAD_TABLE, \p E_BAD_STORAGE_TYPE.
1617 *
1618 * \sa osl_storage_type, osl_open_disk_object().
1619 */
1620 int osl_get_object(const struct osl_table *t, const struct osl_row *r,
1621 unsigned col_num, struct osl_object *object)
1622 {
1623 const struct osl_column_description *cd;
1624
1625 if (!t)
1626 return -E_BAD_TABLE;
1627 cd = get_column_description(t->desc, col_num);
1628 /* col must not be disk storage */
1629 if (cd->storage_type == OSL_DISK_STORAGE)
1630 return -E_BAD_STORAGE_TYPE;
1631 if (cd->storage_type == OSL_MAPPED_STORAGE)
1632 return get_mapped_object(t, col_num, r->id, object);
1633 /* volatile */
1634 *object = r->volatile_objects[t->columns[col_num].volatile_num];
1635 return 1;
1636 }
1637
1638 static int mark_mapped_object_invalid(const struct osl_table *t, uint32_t id,
1639 unsigned col_num)
1640 {
1641 struct osl_object obj;
1642 char *p;
1643 int ret = get_mapped_object(t, col_num, id, &obj);
1644
1645 if (ret < 0)
1646 return ret;
1647 p = obj.data;
1648 p--;
1649 *p = 0xff;
1650 return 1;
1651 }
1652
1653 /**
1654 * Delete a row from an osl table.
1655 *
1656 * \param t Pointer to an open osl table.
1657 * \param row Pointer to the row to delete.
1658 *
1659 * This removes all disk storage objects, removes all rbtree nodes, and frees
1660 * all volatile objects belonging to the given row. For mapped columns, the
1661 * data is merely marked invalid and may be pruned from time to time by
1662 * para_fsck.
1663 *
1664 * \return Positive on success, negative on errors. Possible errors include:
1665 * \p E_BAD_TABLE, errors returned by osl_get_object().
1666 */
1667 int osl_del_row(struct osl_table *t, struct osl_row *row)
1668 {
1669 struct osl_row *r = row;
1670 int i, ret;
1671 const struct osl_column_description *cd;
1672
1673 if (!t)
1674 return -E_BAD_TABLE;
1675 PARA_INFO_LOG("deleting row %p\n", row);
1676
1677 if (t->num_disk_storage_columns) {
1678 char *ds_name;
1679 ret = disk_storage_name_of_row(t, r, &ds_name);
1680 if (ret < 0)
1681 goto out;
1682 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd)
1683 delete_disk_storage_file(t, i, ds_name);
1684 free(ds_name);
1685 }
1686 FOR_EACH_COLUMN(i, t->desc, cd) {
1687 struct osl_column *col = t->columns + i;
1688 enum osl_storage_type st = cd->storage_type;
1689 remove_rb_node(t, i, r);
1690 if (st == OSL_MAPPED_STORAGE) {
1691 mark_mapped_object_invalid(t, r->id, i);
1692 continue;
1693 }
1694 if (st == OSL_NO_STORAGE)
1695 free(r->volatile_objects[col->volatile_num].data);
1696 }
1697 if (t->num_mapped_columns) {
1698 ret = mark_row_invalid(t, r->id);
1699 if (ret < 0)
1700 goto out;
1701 t->num_invalid_rows++;
1702 } else
1703 t->num_rows--;
1704 ret = 1;
1705 out:
1706 free(r->volatile_objects);
1707 free(r);
1708 return ret;
1709 }
1710
1711 /* test if column has an rbtree */
1712 static int check_rbtree_col(const struct osl_table *t, unsigned col_num,
1713 struct osl_column **col)
1714 {
1715 if (!t)
1716 return -E_BAD_TABLE;
1717 if (!(get_column_description(t->desc, col_num)->storage_flags & OSL_RBTREE))
1718 return -E_BAD_STORAGE_FLAGS;
1719 *col = t->columns + col_num;
1720 return 1;
1721 }
1722
1723 /**
1724 * Get the row that contains the given object.
1725 *
1726 * \param t Pointer to an open osl table.
1727 * \param col_num The number of the column to be searched.
1728 * \param obj The object to be looked up.
1729 * \param result Points to the row containing \a obj.
1730 *
1731 * Lookup \a obj in \a t and return the row containing \a obj. The column
1732 * specified by \a col_num must have an associated rbtree.
1733 *
1734 * \return Positive on success, negative on errors. If an error occured, \a
1735 * result is set to \p NULL. Possible errors include: \p E_BAD_TABLE, \p
1736 * E_BAD_STORAGE_FLAGS, errors returned by get_mapped_object(), \p
1737 * E_RB_KEY_NOT_FOUND.
1738 *
1739 * \sa osl_storage_flags
1740 */
1741 int osl_get_row(const struct osl_table *t, unsigned col_num,
1742 const struct osl_object *obj, struct osl_row **result)
1743 {
1744 int ret;
1745 struct rb_node *node;
1746 struct osl_row *row;
1747 struct osl_column *col;
1748
1749 *result = NULL;
1750 ret = check_rbtree_col(t, col_num, &col);
1751 if (ret < 0)
1752 return ret;
1753 ret = search_rbtree(obj, t, col_num, &node, NULL);
1754 if (ret < 0)
1755 return ret;
1756 row = get_row_pointer(node, t->columns[col_num].rbtree_num);
1757 *result = row;
1758 return 1;
1759 }
1760
1761 static int rbtree_loop(struct osl_column *col, void *private_data,
1762 osl_rbtree_loop_func *func)
1763 {
1764 struct rb_node *n;
1765
1766 for (n = rb_first(&col->rbtree); n; n = rb_next(n)) {
1767 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1768 int ret = func(r, private_data);
1769 if (ret < 0)
1770 return ret;
1771 }
1772 return 1;
1773 }
1774
1775 static int rbtree_loop_reverse(struct osl_column *col, void *private_data,
1776 osl_rbtree_loop_func *func)
1777 {
1778 struct rb_node *n;
1779
1780 for (n = rb_last(&col->rbtree); n; n = rb_prev(n)) {
1781 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1782 int ret = func(r, private_data);
1783 if (ret < 0)
1784 return ret;
1785 }
1786 return 1;
1787 }
1788
1789 /**
1790 * Loop over all nodes in an rbtree.
1791 *
1792 * \param t Pointer to an open osl table.
1793 * \param col_num The column to use for iterating over the elements.
1794 * \param private_data Pointer that gets passed to \a func.
1795 * \param func The function to be called for each node in the rbtree.
1796 *
1797 * This function does an in-order walk of the rbtree associated with \a
1798 * col_num. It is an error if the \p OSL_RBTREE flag is not set for this
1799 * column. For each node in the rbtree, the given function \a func is called
1800 * with two \p void* pointers as arguments: The first argument points to the
1801 * row that contains the object corresponding to the rbtree node currently
1802 * traversed, and the \a private_data pointer is passed to \a func as the
1803 * second argument. The loop terminates either if \a func returns a negative
1804 * value, or if all nodes of the tree have been visited.
1805 *
1806 *
1807 * \return Positive on success, negative on errors. If the termination of the
1808 * loop was caused by \a func returning a negative value, this value is
1809 * returned.
1810 *
1811 * \sa osl_storage_flags, osl_rbtree_loop_reverse(), osl_compare_func.
1812 */
1813 int osl_rbtree_loop(const struct osl_table *t, unsigned col_num,
1814 void *private_data, osl_rbtree_loop_func *func)
1815 {
1816 struct osl_column *col;
1817
1818 int ret = check_rbtree_col(t, col_num, &col);
1819 if (ret < 0)
1820 return ret;
1821 return rbtree_loop(col, private_data, func);
1822 }
1823
1824 /**
1825 * Loop over all nodes in an rbtree in reverse order.
1826 *
1827 * \param t Identical meaning as in \p osl_rbtree_loop().
1828 * \param col_num Identical meaning as in \p osl_rbtree_loop().
1829 * \param private_data Identical meaning as in \p osl_rbtree_loop().
1830 * \param func Identical meaning as in \p osl_rbtree_loop().
1831 *
1832 * This function is identical to \p osl_rbtree_loop(), the only difference
1833 * is that the tree is walked in reverse order.
1834 *
1835 * \return The same return value as \p osl_rbtree_loop().
1836 *
1837 * \sa osl_rbtree_loop().
1838 */
1839 int osl_rbtree_loop_reverse(const struct osl_table *t, unsigned col_num,
1840 void *private_data, osl_rbtree_loop_func *func)
1841 {
1842 struct osl_column *col;
1843
1844 int ret = check_rbtree_col(t, col_num, &col);
1845 if (ret < 0)
1846 return ret;
1847 return rbtree_loop_reverse(col, private_data, func);
1848 }
1849
1850 /* TODO: Rollback changes on errors */
1851 static int rename_disk_storage_objects(struct osl_table *t,
1852 struct osl_object *old_obj, struct osl_object *new_obj)
1853 {
1854 int i, ret;
1855 const struct osl_column_description *cd;
1856 char *old_ds_name, *new_ds_name;
1857
1858 if (!t->num_disk_storage_columns)
1859 return 1; /* nothing to do */
1860 if (old_obj->size == new_obj->size && !memcmp(new_obj->data,
1861 old_obj->data, new_obj->size))
1862 return 1; /* object did not change */
1863 old_ds_name = disk_storage_name_of_object(t, old_obj);
1864 new_ds_name = disk_storage_name_of_object(t, new_obj);
1865 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1866 char *old_filename, *new_filename;
1867 ret = create_disk_storage_object_dir(t, i, new_ds_name);
1868 if (ret < 0)
1869 goto out;
1870 old_filename = disk_storage_path(t, i, old_ds_name);
1871 new_filename = disk_storage_path(t, i, new_ds_name);
1872 ret = para_rename(old_filename, new_filename);
1873 free(old_filename);
1874 free(new_filename);
1875 if (ret < 0)
1876 goto out;
1877 }
1878 ret = 1;
1879 out:
1880 free(old_ds_name);
1881 free(new_ds_name);
1882 return ret;
1883
1884 }
1885
1886 /**
1887 * Change an object in an osl table.
1888 *
1889 * \param t Pointer to an open osl table.
1890 * \param r Pointer to the row containing the object to be updated.
1891 * \param col_num Number of the column containing the object to be updated.
1892 * \param obj Pointer to the replacement object.
1893 *
1894 * This function gets rid of all references to the old object. This includes
1895 * removal of the rbtree node in case there is an rbtree associated with \a
1896 * col_num. It then inserts \a obj into the table and the rbtree if neccessary.
1897 *
1898 * If the \p OSL_RBTREE flag is set for \a col_num, you \b MUST call this
1899 * function in order to change the contents of an object, even for volatile or
1900 * mapped columns of constant size (which may be updated directly if \p
1901 * OSL_RBTREE is not set). Otherwise the rbtree might become corrupted.
1902 *
1903 * \return Positive on success, negative on errors. Possible errors include: \p
1904 * E_BAD_TABLE, \p E_RB_KEY_EXISTS, \p E_BAD_SIZE, \p E_NOENT, \p E_UNLINK,
1905 * errors returned by para_write_file(), \p E_MKDIR.
1906 */
1907 int osl_update_object(struct osl_table *t, const struct osl_row *r,
1908 unsigned col_num, struct osl_object *obj)
1909 {
1910 struct osl_column *col;
1911 const struct osl_column_description *cd;
1912 int ret;
1913
1914 if (!t)
1915 return -E_BAD_TABLE;
1916 col = &t->columns[col_num];
1917 cd = get_column_description(t->desc, col_num);
1918 if (cd->storage_flags & OSL_RBTREE) {
1919 if (search_rbtree(obj, t, col_num, NULL, NULL) > 0)
1920 return -E_RB_KEY_EXISTS;
1921 }
1922 if (cd->storage_flags & OSL_FIXED_SIZE) {
1923 if (obj->size != cd->data_size)
1924 return -E_BAD_SIZE;
1925 }
1926 remove_rb_node(t, col_num, r);
1927 if (cd->storage_type == OSL_NO_STORAGE) { /* TODO: If fixed size, reuse object? */
1928 free(r->volatile_objects[col->volatile_num].data);
1929 r->volatile_objects[col->volatile_num] = *obj;
1930 } else if (cd->storage_type == OSL_DISK_STORAGE) {
1931 char *ds_name;
1932 ret = disk_storage_name_of_row(t, r, &ds_name);
1933 if (ret < 0)
1934 return ret;
1935 ret = delete_disk_storage_file(t, col_num, ds_name);
1936 if (ret < 0 && ret != -E_NOENT) {
1937 free(ds_name);
1938 return ret;
1939 }
1940 ret = write_disk_storage_file(t, col_num, obj, ds_name);
1941 free(ds_name);
1942 if (ret < 0)
1943 return ret;
1944 } else { /* mapped storage */
1945 struct osl_object old_obj;
1946 ret = get_mapped_object(t, col_num, r->id, &old_obj);
1947 if (ret < 0)
1948 return ret;
1949 /*
1950 * If the updated column is the disk storage name column, the
1951 * disk storage name changes, so we have to rename all disk
1952 * storage objects accordingly.
1953 */
1954 if (col_num == t->disk_storage_name_column) {
1955 ret = rename_disk_storage_objects(t, &old_obj, obj);
1956 if (ret < 0)
1957 return ret;
1958 }
1959 if (cd->storage_flags & OSL_FIXED_SIZE)
1960 memcpy(old_obj.data, obj->data, cd->data_size);
1961 else { /* TODO: if the size doesn't change, use old space */
1962 uint32_t new_data_map_size;
1963 char *index_entry;
1964 ret = get_index_entry_start(t, r->id, &index_entry);
1965 if (ret < 0)
1966 return ret;
1967 ret = mark_mapped_object_invalid(t, r->id, col_num);
1968 if (ret < 0)
1969 return ret;
1970 ret = append_map_file(t, col_num, obj,
1971 &new_data_map_size);
1972 if (ret < 0)
1973 return ret;
1974 update_index_entry(index_entry, col, new_data_map_size,
1975 obj->size);
1976 }
1977 }
1978 if (cd->storage_flags & OSL_RBTREE) {
1979 ret = insert_rbtree(t, col_num, r, obj);
1980 if (ret < 0)
1981 return ret;
1982 }
1983 return 1;
1984 }
1985
1986 /**
1987 * Retrieve an object of type \p OSL_DISK_STORAGE by row and column.
1988 *
1989 * \param t Pointer to an open osl table.
1990 * \param r Pointer to the row containing the object.
1991 * \param col_num The column number.
1992 * \param obj Points to the result upon successful return.
1993 *
1994 * For columns of type \p OSL_DISK_STORAGE, this function must be used to
1995 * retrieve one of its containing objects. Afterwards, osl_close_disk_object()
1996 * must be called in order to deallocate the resources.
1997 *
1998 * \return Positive on success, negative on errors. Possible errors include:
1999 * \p E_BAD_TABLE, \p E_BAD_STORAGE_TYPE, errors returned by osl_get_object().
2000 *
2001 * \sa osl_get_object(), osl_storage_type, osl_close_disk_object().
2002 */
2003 int osl_open_disk_object(const struct osl_table *t, const struct osl_row *r,
2004 unsigned col_num, struct osl_object *obj)
2005 {
2006 const struct osl_column_description *cd;
2007 char *ds_name, *filename;
2008 int ret;
2009
2010 if (!t)
2011 return -E_BAD_TABLE;
2012 cd = get_column_description(t->desc, col_num);
2013 if (cd->storage_type != OSL_DISK_STORAGE)
2014 return -E_BAD_STORAGE_TYPE;
2015
2016 ret = disk_storage_name_of_row(t, r, &ds_name);
2017 if (ret < 0)
2018 return ret;
2019 filename = disk_storage_path(t, col_num, ds_name);
2020 free(ds_name);
2021 PARA_DEBUG_LOG("filename: %s\n", filename);
2022 ret = mmap_full_file(filename, O_RDONLY, obj);
2023 free(filename);
2024 return ret;
2025 }
2026
2027 /**
2028 * Free resources that were allocated during osl_open_disk_object().
2029 *
2030 * \param obj Pointer to the object previously returned by open_disk_object().
2031 *
2032 * \return The return value of the underlying call to para_munmap().
2033 *
2034 * \sa para_munmap().
2035 */
2036 int osl_close_disk_object(struct osl_object *obj)
2037 {
2038 return para_munmap(obj->data, obj->size);
2039 }
2040
2041 /**
2042 * Get the number of rows of the given table.
2043 *
2044 * \param t Pointer to an open osl table.
2045 * \param num_rows Result is returned here.
2046 *
2047 * The number of rows returned via \a num_rows excluding any invalid rows.
2048 *
2049 * \return Positive on success, \p -E_BAD_TABLE if \a t is \p NULL.
2050 */
2051 int osl_get_num_rows(const struct osl_table *t, unsigned *num_rows)
2052 {
2053 if (!t)
2054 return -E_BAD_TABLE;
2055 assert(t->num_rows >= t->num_invalid_rows);
2056 *num_rows = t->num_rows - t->num_invalid_rows;
2057 return 1;
2058 }
2059
2060 /**
2061 * Get the rank of a row.
2062 *
2063 * \param t An open osl table.
2064 * \param r The row to get the rank of.
2065 * \param col_num The number of an rbtree column.
2066 * \param rank Result pointer.
2067 *
2068 * The rank is, by definition, the position of the row in the linear order
2069 * determined by an inorder tree walk of the rbtree associated with column
2070 * number \a col_num of \a table.
2071 *
2072 * \return Positive on success, negative on errors.
2073 *
2074 * \sa osl_get_nth_row().
2075 */
2076 int osl_get_rank(const struct osl_table *t, struct osl_row *r,
2077 unsigned col_num, unsigned *rank)
2078 {
2079 struct osl_object obj;
2080 struct osl_column *col;
2081 struct rb_node *node;
2082 int ret = check_rbtree_col(t, col_num, &col);
2083
2084 if (ret < 0)
2085 return ret;
2086 ret = osl_get_object(t, r, col_num, &obj);
2087 if (ret < 0)
2088 return ret;
2089 ret = search_rbtree(&obj, t, col_num, &node, NULL);
2090 if (ret < 0)
2091 return ret;
2092 ret = rb_rank(node, rank);
2093 if (ret < 0)
2094 return -E_BAD_ROW;
2095 return 1;
2096 }
2097
2098 /**
2099 * Get the row with n-th greatest value.
2100 *
2101 * \param t Pointer to an open osl table.
2102 * \param col_num The column number.
2103 * \param n The rank of the desired row.
2104 * \param result Row is returned here.
2105 *
2106 * Retrieve the n-th order statistic with respect to the compare function
2107 * of the rbtree column \a col_num. In other words, get that row with
2108 * \a n th greatest value in column \a col_num. It's an error if
2109 * \a col_num is not a rbtree column, or if \a n is larger than the
2110 * number of rows in the table.
2111 *
2112 * \return Positive on success, negative on errors. Possible errors:
2113 * \p E_BAD_TABLE, \p E_BAD_STORAGE_FLAGS, \p E_RB_KEY_NOT_FOUND.
2114 *
2115 * \sa osl_storage_flags, osl_compare_func, osl_get_row(),
2116 * osl_rbtree_last_row(), osl_rbtree_first_row(), osl_get_rank().
2117 */
2118 int osl_get_nth_row(const struct osl_table *t, unsigned col_num,
2119 unsigned n, struct osl_row **result)
2120 {
2121 struct osl_column *col;
2122 struct rb_node *node;
2123 int ret = check_rbtree_col(t, col_num, &col);
2124
2125 if (ret < 0)
2126 return ret;
2127 node = rb_nth(col->rbtree.rb_node, n);
2128 if (!node)
2129 return -E_RB_KEY_NOT_FOUND;
2130 *result = get_row_pointer(node, col->rbtree_num);
2131 return 1;
2132 }
2133
2134 /**
2135 * Get the row corresponding to the smallest rbtree node of a column.
2136 *
2137 * \param t An open rbtree table.
2138 * \param col_num The number of the rbtree column.
2139 * \param result A pointer to the first row is returned here.
2140 *
2141 * The rbtree node of the smallest object (with respect to the corresponding
2142 * compare function) is selected and the row containing this object is
2143 * returned. It is an error if \a col_num refers to a column without an
2144 * associated rbtree.
2145 *
2146 * \return Positive on success, negative on errors.
2147 *
2148 * \sa osl_get_nth_row(), osl_rbtree_last_row().
2149 */
2150 int osl_rbtree_first_row(const struct osl_table *t, unsigned col_num,
2151 struct osl_row **result)
2152 {
2153 return osl_get_nth_row(t, col_num, 1, result);
2154 }
2155
2156 /**
2157 * Get the row corresponding to the greatest rbtree node of a column.
2158 *
2159 * \param t The same meaning as in \p osl_rbtree_first_row().
2160 * \param col_num The same meaning as in \p osl_rbtree_first_row().
2161 * \param result The same meaning as in \p osl_rbtree_first_row().
2162 *
2163 * This function works just like osl_rbtree_first_row(), the only difference
2164 * is that the row containing the greatest rather than the smallest object is
2165 * returned.
2166 *
2167 * \return Positive on success, negative on errors.
2168 *
2169 * \sa osl_get_nth_row(), osl_rbtree_first_row().
2170 */
2171 int osl_rbtree_last_row(const struct osl_table *t, unsigned col_num,
2172 struct osl_row **result)
2173 {
2174 unsigned num_rows;
2175 int ret = osl_get_num_rows(t, &num_rows);
2176
2177 if (ret < 0)
2178 return ret;
2179 return osl_get_nth_row(t, col_num, num_rows, result);
2180 }