2 * Copyright (C) 2007-2009 Andre Noll <maan@tuebingen.mpg.de>
4 * Licensed under the GPL v2. For licencing details see COPYING.
7 /** \file osl.c Object storage layer functions. */
8 #include <dirent.h> /* readdir() */
16 /* Taken from Drepper: How to write shared libraries, Appendix B. */
18 #define MSGSTRFIELD(line) MSGSTRFIELD1(line)
19 #define MSGSTRFIELD1(line) str##line
20 static const union msgstr_t {
22 #define OSL_ERROR(n, s) char MSGSTRFIELD(__LINE__)[sizeof(s)];
28 #define OSL_ERROR(n, s) s,
32 static const unsigned int errmsgidx[] = {
33 #define OSL_ERROR(n, s) [n] = offsetof(union msgstr_t, MSGSTRFIELD(__LINE__)),
38 __export const char *osl_strerror(int num)
40 return msgstr.str + errmsgidx[num];
45 static void __attribute ((constructor)) init_loglevel(void)
47 char *p = getenv("OSL_LOGLEVEL");
49 /* don't log anything if unset */
50 loglevel = p? atoi(p) : EMERG + 1;
57 * \param fmt Usual format string.
59 * All XXX_LOG() macros use this function.
61 __printf_2_3 void __log(int ll, const char* fmt,...)
72 strftime(str, sizeof(str), "%b %d %H:%M:%S", tm);
73 fprintf(stderr, "%s ", str);
75 vfprintf(stderr, fmt, argp);
80 * A wrapper for lseek(2).
82 * \param fd The file descriptor whose offset is to be to repositioned.
83 * \param offset A value-result parameter.
84 * \param whence Usual repositioning directive.
86 * Reposition the offset of the file descriptor \a fd to the argument \a offset
87 * according to the directive \a whence. Upon successful return, \a offset
88 * contains the resulting offset location as measured in bytes from the
89 * beginning of the file.
91 * \return Positive on success. Otherwise, the function returns \p -E_OSL_LSEEK.
95 static int __lseek(int fd, off_t *offset, int whence)
97 *offset = lseek(fd, *offset, whence);
98 int ret = -E_OSL_LSEEK;
104 static int append_file(const char *filename, char *data, size_t data_size,
109 // DEBUG_LOG("appending %zu + %zu bytes\n", header_size, data_size);
110 ret = osl_open(filename, O_WRONLY | O_CREAT | O_APPEND, 0644);
114 ret = write_all(fd, data, &data_size);
119 ret = __lseek(fd, &offset, SEEK_END);
122 // DEBUG_LOG("new file size: " FMT_OFF_T "\n", offset);
131 static int verify_name(const char *name)
134 return -E_OSL_BAD_NAME;
136 return -E_OSL_BAD_NAME;
137 if (strchr(name, '/'))
138 return -E_OSL_BAD_NAME;
139 if (!strcmp(name, ".."))
140 return -E_OSL_BAD_NAME;
141 if (!strcmp(name, "."))
142 return -E_OSL_BAD_NAME;
146 static char *disk_storage_dirname(const struct osl_table *t, unsigned col_num,
149 char *dirname, *column_name = column_filename(t, col_num);
153 if (!(t->desc->flags & OSL_LARGE_TABLE))
155 dirname = make_message("%s/%.2s", column_name, ds_name);
160 static char *disk_storage_name_of_object(const struct osl_table *t,
161 const struct osl_object *obj)
163 HASH_TYPE hash[HASH_SIZE];
164 hash_object(obj, hash);
165 return disk_storage_name_of_hash(t, hash);
168 static int disk_storage_name_of_row(const struct osl_table *t,
169 const struct osl_row *row, char **name)
171 struct osl_object obj;
172 int ret = osl_get_object(t, row, t->disk_storage_name_column, &obj);
176 *name = disk_storage_name_of_object(t, &obj);
182 static void column_name_hash(const char *col_name, HASH_TYPE *hash)
184 hash_function(col_name, strlen(col_name), hash);
187 static int init_column_descriptions(struct osl_table *t)
190 const struct osl_column_description *cd;
192 ret = verify_name(t->desc->name);
195 ret = -E_OSL_BAD_DB_DIR;
196 if (!t->desc->dir && (t->num_disk_storage_columns || t->num_mapped_columns))
198 /* the size of the index header without column descriptions */
199 t->index_header_size = IDX_COLUMN_DESCRIPTIONS;
200 FOR_EACH_COLUMN(i, t->desc, cd) {
201 struct osl_column *col = t->columns + i;
202 if (cd->storage_flags & OSL_RBTREE) {
203 if (!cd->compare_function)
204 return -E_OSL_NO_COMPARE_FUNC;
206 if (cd->storage_type == OSL_NO_STORAGE)
208 ret = -E_OSL_NO_COLUMN_NAME;
209 if (!cd->name || !cd->name[0])
211 ret = verify_name(cd->name);
214 t->index_header_size += index_column_description_size(cd->name);
215 column_name_hash(cd->name, col->name_hash);
216 ret = -E_OSL_DUPLICATE_COL_NAME;
217 for (j = i + 1; j < t->desc->num_columns; j++) {
218 const char *name2 = get_column_description(t->desc,
220 if (cd->name && name2 && !strcmp(cd->name, name2))
230 * Initialize a struct table from given table description.
232 * \param desc The description of the osl table.
233 * \param table_ptr Result is returned here.
235 * This function performs several sanity checks on \p desc and returns if any
236 * of these tests fail. On success, a struct \p osl_table is allocated and
237 * initialized with data derived from \p desc.
241 * \sa struct osl_table.
243 int init_table_structure(const struct osl_table_description *desc,
244 struct osl_table **table_ptr)
246 const struct osl_column_description *cd;
247 struct osl_table *t = calloc(1, sizeof(*t));
248 int i, ret = -E_OSL_NOMEM, have_disk_storage_name_column = 0;
252 ret = -E_OSL_BAD_TABLE_DESC;
255 DEBUG_LOG("creating table structure for '%s' from table "
256 "description\n", desc->name);
257 ret = -E_OSL_NO_COLUMN_DESC;
258 if (!desc->column_descriptions)
260 ret = -E_OSL_NO_COLUMNS;
261 if (!desc->num_columns)
264 t->columns = calloc(desc->num_columns, sizeof(struct osl_column));
268 FOR_EACH_COLUMN(i, t->desc, cd) {
269 enum osl_storage_type st = cd->storage_type;
270 enum osl_storage_flags sf = cd->storage_flags;
271 struct osl_column *col = &t->columns[i];
273 ret = -E_OSL_BAD_STORAGE_TYPE;
274 if (st != OSL_MAPPED_STORAGE && st != OSL_DISK_STORAGE
275 && st != OSL_NO_STORAGE)
277 ret = -E_OSL_BAD_STORAGE_FLAGS;
278 if (st == OSL_DISK_STORAGE && sf & OSL_RBTREE)
280 if ((sf & OSL_RBTREE) && !(sf & OSL_UNIQUE))
281 WARNING_LOG("invalid storage flags for column %s: "
282 "OSL_RBTREE && !OSL_UNIQUE\n", cd->name);
283 ret = -E_OSL_BAD_STORAGE_SIZE;
284 if (sf & OSL_FIXED_SIZE && !cd->data_size)
287 case OSL_DISK_STORAGE:
288 t->num_disk_storage_columns++;
290 case OSL_MAPPED_STORAGE:
291 t->num_mapped_columns++;
292 col->index_offset = t->row_index_size;
293 t->row_index_size += 8;
296 col->volatile_num = t->num_volatile_columns;
297 t->num_volatile_columns++;
300 if (sf & OSL_RBTREE) {
301 col->rbtree_num = t->num_rbtrees;
303 if ((sf & OSL_UNIQUE) && (st == OSL_MAPPED_STORAGE)) {
304 if (!have_disk_storage_name_column)
305 t->disk_storage_name_column = i;
306 have_disk_storage_name_column = 1;
310 ret = -E_OSL_NO_UNIQUE_RBTREE_COLUMN;
311 if (t->num_disk_storage_columns && !have_disk_storage_name_column)
313 ret = -E_OSL_NO_RBTREE_COL;
317 DEBUG_LOG("OK. Index entry size: %u\n", t->row_index_size);
318 ret = init_column_descriptions(t);
330 * Read the table description from index header.
332 * \param map The memory mapping of the index file.
333 * \param desc The values found in the index header are returned here.
335 * Read the index header, check for the osl magic string and the table version
336 * number. Read all information stored in the index header into \a desc.
340 * \sa struct osl_table_description, osl_create_table.
342 int read_table_desc(struct osl_object *map, struct osl_table_description *desc)
344 char *buf = map->data;
345 uint8_t version, compat_version, create_version;
346 uint16_t header_size;
349 struct osl_column_description *cd;
351 if (map->size < MIN_INDEX_HEADER_SIZE(1))
352 return -E_OSL_SHORT_TABLE;
353 if (strncmp(buf + IDX_OSL_MAGIC, OSL_MAGIC, strlen(OSL_MAGIC)))
354 return -E_OSL_NO_MAGIC;
355 version = read_u8(buf + IDX_VERSION);
357 * The on-disk version consists of two version numbers: the
358 * create_version (low 4 bits) is the CURRENT_TABLE_VERSION version
359 * number of the library that created the table, and compat_version
360 * (high 4 bits) tells us the lowest version of the library that can
361 * still read this table.
363 create_version = version & 0xf;
364 compat_version = version >> 4;
365 INFO_LOG("create_version: %u, compat_version: %u\n", create_version,
367 if (create_version < MIN_TABLE_VERSION /* table too old */
368 || compat_version > CURRENT_TABLE_VERSION) /* libosl too old */
369 return -E_OSL_VERSION_MISMATCH;
370 desc->flags = read_u8(buf + IDX_TABLE_FLAGS);
371 desc->num_columns = read_u16(buf + IDX_NUM_COLUMNS);
372 INFO_LOG("%u columns\n", desc->num_columns);
373 if (!desc->num_columns)
374 return -E_OSL_NO_COLUMNS;
375 header_size = read_u16(buf + IDX_HEADER_SIZE);
376 if (map->size < header_size)
377 return -E_OSL_BAD_SIZE;
378 desc->column_descriptions = calloc(desc->num_columns,
379 sizeof(struct osl_column_description));
380 if (!desc->column_descriptions)
382 offset = IDX_COLUMN_DESCRIPTIONS;
383 FOR_EACH_COLUMN(i, desc, cd) {
386 ret = -E_OSL_SHORT_TABLE;
387 if (map->size < offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE) {
388 ERROR_LOG("map size = %zu < %u = offset + min desc size\n",
389 map->size, offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE);
392 cd->storage_type = read_u16(buf + offset + IDX_CD_STORAGE_TYPE);
393 cd->storage_flags = read_u16(buf + offset +
394 IDX_CD_STORAGE_FLAGS);
395 cd->data_size = read_u32(buf + offset + IDX_CD_DATA_SIZE);
396 null_byte = memchr(buf + offset + IDX_CD_NAME, '\0',
397 map->size - offset - IDX_CD_NAME);
398 ret = -E_OSL_INDEX_CORRUPTION;
402 cd->name = strdup(buf + offset + IDX_CD_NAME);
405 offset += index_column_description_size(cd->name);
407 if (offset != header_size) {
408 ret = -E_OSL_INDEX_CORRUPTION;
409 ERROR_LOG("real header size = %u != %u = stored header size\n",
410 offset, header_size);
415 FOR_EACH_COLUMN(i, desc, cd)
421 * check whether the table description given by \p t->desc matches the on-disk
422 * table structure stored in the index of \a t.
424 static int compare_table_descriptions(struct osl_table *t)
427 struct osl_table_description desc;
428 const struct osl_column_description *cd1, *cd2;
430 /* read the on-disk structure into desc */
431 ret = read_table_desc(&t->index_map, &desc);
434 ret = -E_OSL_BAD_TABLE_FLAGS;
435 if (desc.flags != t->desc->flags)
437 ret = -E_OSL_BAD_COLUMN_NUM;
438 if (desc.num_columns > t->desc->num_columns)
440 if (desc.num_columns < t->desc->num_columns) {
441 struct osl_column_description *cd;
442 unsigned diff = t->desc->num_columns - desc.num_columns;
443 INFO_LOG("extending table by %u volatile columns\n", diff);
445 desc.column_descriptions = realloc(desc.column_descriptions,
446 t->desc->num_columns * sizeof(struct osl_column_description));
447 if (!desc.column_descriptions)
449 for (i = desc.num_columns; i < t->desc->num_columns; i++) {
450 cd = get_column_description(&desc, i);
451 cd->storage_type = OSL_NO_STORAGE;
454 desc.num_columns += diff;
456 FOR_EACH_COLUMN(i, t->desc, cd1) {
457 cd2 = get_column_description(&desc, i);
458 ret = -E_OSL_BAD_STORAGE_TYPE;
459 if (cd1->storage_type != cd2->storage_type)
461 if (cd1->storage_type == OSL_NO_STORAGE)
463 ret = -E_OSL_BAD_STORAGE_FLAGS;
464 if (cd1->storage_flags != cd2->storage_flags) {
465 ERROR_LOG("sf1 = %u != %u = sf2\n",
466 cd1->storage_flags, cd2->storage_flags);
469 ret = -E_OSL_BAD_DATA_SIZE;
470 if (cd1->storage_flags & OSL_FIXED_SIZE)
471 if (cd1->data_size != cd2->data_size)
473 ret = -E_OSL_BAD_COLUMN_NAME;
474 if (strcmp(cd1->name, cd2->name))
477 INFO_LOG("table description of '%s' matches on-disk data, good\n",
481 FOR_EACH_COLUMN(i, &desc, cd1)
483 free(desc.column_descriptions);
487 static int create_table_index(struct osl_table *t)
489 char *buf, *filename;
491 size_t size = t->index_header_size;
492 const struct osl_column_description *cd;
495 INFO_LOG("creating %zu byte index for table %s\n", size,
497 buf = calloc(1, size);
500 sprintf(buf + IDX_OSL_MAGIC, "%s", OSL_MAGIC);
501 write_u8(buf + IDX_TABLE_FLAGS, t->desc->flags);
502 write_u8(buf + IDX_DIRTY_FLAG, 0);
503 write_u8(buf + IDX_VERSION, CURRENT_TABLE_VERSION
504 + (COMPAT_TABLE_VERSION << 4));
505 write_u16(buf + IDX_NUM_COLUMNS, t->num_mapped_columns + t->num_disk_storage_columns);
506 write_u16(buf + IDX_HEADER_SIZE, t->index_header_size);
507 offset = IDX_COLUMN_DESCRIPTIONS;
508 FOR_EACH_COLUMN(i, t->desc, cd) {
509 /* no need to store info about volatile storage */
510 if (cd->storage_type == OSL_NO_STORAGE)
512 write_u16(buf + offset + IDX_CD_STORAGE_TYPE,
514 write_u16(buf + offset + IDX_CD_STORAGE_FLAGS,
516 if (cd->storage_flags & OSL_FIXED_SIZE)
517 write_u32(buf + offset + IDX_CD_DATA_SIZE,
519 strcpy(buf + offset + IDX_CD_NAME, cd->name);
520 offset += index_column_description_size(cd->name);
522 assert(offset == size);
523 filename = index_filename(t->desc);
525 ret = write_file(filename, buf, size);
533 __export int osl_create_table(const struct osl_table_description *desc)
535 const struct osl_column_description *cd;
536 char *table_dir = NULL, *filename;
538 int i, ret = init_table_structure(desc, &t);
542 INFO_LOG("creating %s\n", desc->name);
543 FOR_EACH_COLUMN(i, t->desc, cd) {
544 if (cd->storage_type == OSL_NO_STORAGE)
547 ret = osl_mkdir(desc->dir, 0777);
548 if (ret < 0 && ret != -E_OSL_DIR_EXISTS)
550 table_dir = make_message("%s/%s", desc->dir,
555 ret = osl_mkdir(table_dir, 0777);
560 filename = column_filename(t, i);
563 INFO_LOG("filename: %s\n", filename);
564 if (cd->storage_type == OSL_MAPPED_STORAGE) {
565 ret = osl_open(filename, O_RDWR | O_CREAT | O_EXCL,
574 ret = osl_mkdir(filename, 0777);
579 if (t->num_mapped_columns) {
580 ret = create_table_index(t);
592 static int table_is_dirty(struct osl_table *t)
594 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
595 uint8_t dirty = read_u8(buf) & 0x1;
599 static void mark_table_dirty(struct osl_table *t)
601 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
602 write_u8(buf, read_u8(buf) | 1);
605 static void mark_table_clean(struct osl_table *t)
607 char *buf = (char *)t->index_map.data + IDX_DIRTY_FLAG;
608 write_u8(buf, read_u8(buf) & 0xfe);
611 static void unmap_column(struct osl_table *t, unsigned col_num)
613 struct osl_object map = t->columns[col_num].data_map;
617 ret = osl_munmap(map.data, map.size);
623 * Unmap all mapped files of an osl table.
625 * \param t Pointer to a mapped table.
626 * \param flags Options for unmapping.
628 * \return Positive on success, negative on errors.
630 * \sa map_table(), enum osl_close_flags, osl_munmap().
632 int unmap_table(struct osl_table *t, enum osl_close_flags flags)
635 const struct osl_column_description *cd;
638 if (!t->num_mapped_columns) /* can this ever happen? */
640 INFO_LOG("unmapping table '%s'\n", t->desc->name);
641 if (!t->index_map.data)
642 return -E_OSL_NOT_MAPPED;
643 if (flags & OSL_MARK_CLEAN)
645 ret = osl_munmap(t->index_map.data, t->index_map.size);
648 t->index_map.data = NULL;
651 FOR_EACH_MAPPED_COLUMN(i, t, cd)
656 static int map_column(struct osl_table *t, unsigned col_num)
659 char *filename = column_filename(t, col_num);
664 ret = osl_stat(filename, &statbuf);
669 if (!(S_IFREG & statbuf.st_mode)) {
673 ret = mmap_full_file(filename, O_RDWR,
674 &t->columns[col_num].data_map.data,
675 &t->columns[col_num].data_map.size,
682 * Map the index file and all columns of type \p OSL_MAPPED_STORAGE into memory.
684 * \param t Pointer to an initialized table structure.
685 * \param flags Mapping options.
687 * \return Negative return value on errors; on success the number of rows
688 * (including invalid rows) is returned.
690 * \sa unmap_table(), enum map_table_flags, osl_open_table(), mmap(2).
692 int map_table(struct osl_table *t, enum map_table_flags flags)
695 const struct osl_column_description *cd;
696 int i = 0, ret, num_rows = 0;
698 if (!t->num_mapped_columns)
700 if (t->index_map.data)
701 return -E_OSL_ALREADY_MAPPED;
702 filename = index_filename(t->desc);
705 INFO_LOG("mapping table '%s' (index: %s)\n", t->desc->name, filename);
706 ret = mmap_full_file(filename, flags & MAP_TBL_FL_MAP_RDONLY?
707 O_RDONLY : O_RDWR, &t->index_map.data, &t->index_map.size, NULL);
711 if (flags & MAP_TBL_FL_VERIFY_INDEX) {
712 ret = compare_table_descriptions(t);
717 if (!(flags & MAP_TBL_FL_IGNORE_DIRTY)) {
718 if (table_is_dirty(t)) {
719 ERROR_LOG("%s is dirty\n", t->desc->name);
724 num_rows = table_num_rows(t);
728 FOR_EACH_MAPPED_COLUMN(i, t, cd) {
729 ret = map_column(t, i);
734 err: /* unmap what is already mapped */
735 for (i--; i >= 0; i--) {
736 struct osl_object map = t->columns[i].data_map;
737 osl_munmap(map.data, map.size);
740 osl_munmap(t->index_map.data, t->index_map.size);
741 t->index_map.data = NULL;
746 * Retrieve a mapped object by row and column number.
748 * \param t Pointer to an open osl table.
749 * \param col_num Number of the mapped column containing the object to retrieve.
750 * \param row_num Number of the row containing the object to retrieve.
751 * \param obj The result is returned here.
753 * It is considered an error if \a col_num does not refer to a column
754 * of storage type \p OSL_MAPPED_STORAGE.
758 * \sa osl_storage_type.
760 int get_mapped_object(const struct osl_table *t, unsigned col_num,
761 uint32_t row_num, struct osl_object *obj)
763 struct osl_column *col = &t->columns[col_num];
768 if (t->num_rows <= row_num)
769 return -E_OSL_BAD_ROW_NUM;
770 ret = get_cell_index(t, row_num, col_num, &cell_index);
773 offset = read_u32(cell_index);
774 obj->size = read_u32(cell_index + 4);
775 obj->data = col->data_map.data + offset;
780 * It's OK to call this with result = rb_node = NULL. If result is not NULL,
781 * and rb key was not found, result points to the parent node.
783 static int search_rbtree(const struct osl_object *obj,
784 const struct osl_table *t, unsigned col_num,
785 struct rb_node **result, struct rb_node ***rb_link)
787 struct osl_column *col = &t->columns[col_num];
788 struct rb_node **new = &col->rbtree.rb_node, *parent = NULL;
789 const struct osl_column_description *cd =
790 get_column_description(t->desc, col_num);
791 enum osl_storage_type st = cd->storage_type;
793 struct osl_row *this_row = get_row_pointer(*new,
796 struct osl_object this_obj;
798 if (st == OSL_MAPPED_STORAGE) {
799 ret = get_mapped_object(t, col_num, this_row->num,
804 this_obj = this_row->volatile_objects[col->volatile_num];
805 ret = cd->compare_function(obj, &this_obj);
808 *result = get_rb_node_pointer(this_row,
813 new = &((*new)->rb_left);
815 new = &((*new)->rb_right);
821 return -E_OSL_RB_KEY_NOT_FOUND;
824 static int insert_rbtree(struct osl_table *t, unsigned col_num,
825 const struct osl_row *row, const struct osl_object *obj)
827 struct rb_node *parent, **rb_link;
830 int ret = search_rbtree(obj, t, col_num, &parent, &rb_link);
833 return -E_OSL_RB_KEY_EXISTS;
834 rbtree_num = t->columns[col_num].rbtree_num;
835 n = get_rb_node_pointer(row, rbtree_num);
836 rb_link_node(n, parent, rb_link);
837 rb_insert_color(n, &t->columns[col_num].rbtree);
841 static void remove_rb_node(struct osl_table *t, unsigned col_num,
842 const struct osl_row *row)
844 struct osl_column *col = &t->columns[col_num];
845 const struct osl_column_description *cd =
846 get_column_description(t->desc, col_num);
847 enum osl_storage_flags sf = cd->storage_flags;
848 struct rb_node *victim, *splice_out_node, *tmp;
849 if (!(sf & OSL_RBTREE))
852 * Which node is removed/spliced out actually depends on how many
853 * children the victim node has: If it has no children, it gets
854 * deleted. If it has one child, it gets spliced out. If it has two
855 * children, its successor (which has at most a right child) gets
858 victim = get_rb_node_pointer(row, col->rbtree_num);
859 if (victim->rb_left && victim->rb_right)
860 splice_out_node = rb_next(victim);
862 splice_out_node = victim;
863 /* Go up to the root and decrement the size of each node in the path. */
864 for (tmp = splice_out_node; tmp; tmp = rb_parent(tmp))
866 rb_erase(victim, &col->rbtree);
869 static int add_row_to_rbtrees(struct osl_table *t, uint32_t row_num,
870 struct osl_object *volatile_objs, struct osl_row **row_ptr)
874 struct osl_row *row = allocate_row(t->num_rbtrees);
875 const struct osl_column_description *cd;
880 row->volatile_objects = volatile_objs;
881 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
882 if (cd->storage_type == OSL_MAPPED_STORAGE) {
883 struct osl_object obj;
884 ret = get_mapped_object(t, i, row_num, &obj);
887 ret = insert_rbtree(t, i, row, &obj);
888 } else { /* volatile */
889 const struct osl_object *obj
890 = volatile_objs + t->columns[i].volatile_num;
891 ret = insert_rbtree(t, i, row, obj);
899 err: /* rollback changes, i.e. remove added entries from rbtrees */
901 remove_rb_node(t, i--, row);
906 static void free_volatile_objects(const struct osl_table *t,
907 enum osl_close_flags flags)
911 struct osl_column *rb_col;
912 const struct osl_column_description *cd;
914 if (!t->num_volatile_columns)
916 /* find the first rbtree column (any will do) */
917 FOR_EACH_RBTREE_COLUMN(i, t, cd)
919 rb_col = t->columns + i;
920 /* walk that rbtree and free all volatile objects */
921 for (n = rb_first(&rb_col->rbtree); n; n = rb_next(n)) {
922 struct osl_row *r = get_row_pointer(n, rb_col->rbtree_num);
923 if (flags & OSL_FREE_VOLATILE)
924 FOR_EACH_VOLATILE_COLUMN(j, t, cd) {
925 if (cd->storage_flags & OSL_DONT_FREE)
927 free(r->volatile_objects[
928 t->columns[j].volatile_num].data);
930 // for (j = 0; j < t->num_volatile_columns; j++)
931 // free(r->volatile_objects[j].data);
932 free(r->volatile_objects);
937 * Erase all rbtree nodes and free resources.
939 * \param t Pointer to an open osl table.
941 * This function is called by osl_close_table().
943 void clear_rbtrees(struct osl_table *t)
945 const struct osl_column_description *cd;
946 unsigned i, rbtrees_cleared = 0;
948 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
949 struct osl_column *col = &t->columns[i];
952 for (n = rb_first(&col->rbtree); n;) {
954 rb_erase(n, &col->rbtree);
955 if (rbtrees_cleared == t->num_rbtrees) {
956 r = get_row_pointer(n, col->rbtree_num);
966 __export int osl_close_table(struct osl_table *t, enum osl_close_flags flags)
971 return -E_OSL_BAD_TABLE;
972 NOTICE_LOG("closing table %s\n", t->desc->name);
973 free_volatile_objects(t, flags);
975 ret = unmap_table(t, flags);
977 ERROR_LOG("unmap_table failed: %d\n", ret);
984 * Find out whether the given row number corresponds to an invalid row.
986 * \param t Pointer to the osl table.
987 * \param row_num The number of the row in question.
989 * By definition, a row is considered invalid if all its index entries
992 * \return Positive if \a row_num corresponds to an invalid row,
993 * zero if it corresponds to a valid row, negative on errors.
995 int row_is_invalid(struct osl_table *t, uint32_t row_num)
998 int i, ret = get_row_index(t, row_num, &row_index);
1002 for (i = 0; i < t->row_index_size; i++) {
1003 if ((unsigned char)row_index[i] != 0xff)
1006 INFO_LOG("row %d is invalid\n", row_num);
1011 * Invalidate a row of an osl table.
1013 * \param t Pointer to an open osl table.
1014 * \param row_num Number of the row to mark as invalid.
1016 * This function marks each mapped object in the index entry of \a row as
1021 int mark_row_invalid(struct osl_table *t, uint32_t row_num)
1024 int ret = get_row_index(t, row_num, &row_index);
1028 INFO_LOG("marking row %d as invalid\n", row_num);
1029 memset(row_index, 0xff, t->row_index_size);
1034 * Initialize all rbtrees and compute number of invalid rows.
1036 * \param t The table containing the rbtrees to be initialized.
1040 int init_rbtrees(struct osl_table *t)
1043 const struct osl_column_description *cd;
1045 /* create rbtrees */
1046 FOR_EACH_RBTREE_COLUMN(i, t, cd)
1047 t->columns[i].rbtree = RB_ROOT;
1048 /* add valid rows to rbtrees */
1049 t->num_invalid_rows = 0;
1050 for (i = 0; i < t->num_rows; i++) {
1051 struct osl_object *volatile_objs;
1052 ret = row_is_invalid(t, i);
1056 t->num_invalid_rows++;
1059 if (t->num_volatile_columns > 0) {
1060 volatile_objs = calloc(t->num_volatile_columns,
1061 sizeof(struct osl_object));
1063 return -E_OSL_NOMEM;
1065 volatile_objs = NULL;
1066 ret = add_row_to_rbtrees(t, i, volatile_objs, NULL);
1073 __export int osl_open_table(const struct osl_table_description *table_desc,
1074 struct osl_table **result)
1077 struct osl_table *t;
1078 const struct osl_column_description *cd;
1080 NOTICE_LOG("opening table %s\n", table_desc->name);
1081 ret = init_table_structure(table_desc, &t);
1084 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1085 struct stat statbuf;
1086 char *dirname = column_filename(t, i);
1091 /* check if directory exists */
1092 ret = osl_stat(dirname, &statbuf);
1096 ret = -E_OSL_NOTDIR;
1097 if (!S_ISDIR(statbuf.st_mode))
1100 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1104 DEBUG_LOG("num rows: %d\n", t->num_rows);
1105 ret = init_rbtrees(t);
1107 osl_close_table(t, OSL_MARK_CLEAN); /* ignore further errors */
1118 static int create_disk_storage_object_dir(const struct osl_table *t,
1119 unsigned col_num, const char *ds_name)
1124 if (!(t->desc->flags & OSL_LARGE_TABLE))
1126 dirname = disk_storage_dirname(t, col_num, ds_name);
1128 return -E_OSL_NOMEM;
1129 ret = osl_mkdir(dirname, 0777);
1131 if (ret < 0 && ret != -E_OSL_DIR_EXISTS)
1136 static int write_disk_storage_file(const struct osl_table *t, unsigned col_num,
1137 const struct osl_object *obj, const char *ds_name)
1142 ret = create_disk_storage_object_dir(t, col_num, ds_name);
1145 filename = disk_storage_path(t, col_num, ds_name);
1147 return -E_OSL_NOMEM;
1148 ret = write_file(filename, obj->data, obj->size);
1153 static int append_map_file(const struct osl_table *t, unsigned col_num,
1154 const struct osl_object *obj, uint32_t *new_size)
1156 char *filename = column_filename(t, col_num);
1160 return -E_OSL_NOMEM;
1161 ret = append_file(filename, obj->data, obj->size, new_size);
1166 static int append_row_index(const struct osl_table *t, char *row_index)
1171 if (!t->num_mapped_columns)
1173 filename = index_filename(t->desc);
1175 return -E_OSL_NOMEM;
1176 ret = append_file(filename, row_index, t->row_index_size, NULL);
1181 static int truncate_mapped_file(const struct osl_table *t, unsigned col_num,
1185 char *filename = column_filename(t, col_num);
1188 return -E_OSL_NOMEM;
1189 ret = truncate_file(filename, size);
1194 static int delete_disk_storage_file(const struct osl_table *t, unsigned col_num,
1195 const char *ds_name)
1197 char *dirname, *filename = disk_storage_path(t, col_num, ds_name);
1201 return -E_OSL_NOMEM;
1202 if (unlink(filename) < 0)
1203 ret = errno == ENOENT? -E_OSL_NOENT : -E_OSL_UNLINK;
1207 if (!(t->desc->flags & OSL_LARGE_TABLE))
1209 dirname = disk_storage_dirname(t, col_num, ds_name);
1211 return -E_OSL_NOMEM;
1217 __export int osl_add_and_get_row(struct osl_table *t, struct osl_object *objects,
1218 struct osl_row **row)
1221 char *ds_name = NULL;
1222 struct rb_node **rb_parents = NULL, ***rb_links = NULL;
1223 char *new_row_index = NULL;
1224 struct osl_object *volatile_objs = NULL;
1225 const struct osl_column_description *cd;
1228 return -E_OSL_BAD_TABLE;
1229 rb_parents = malloc(t->num_rbtrees * sizeof(struct rn_node*));
1231 return -E_OSL_NOMEM;
1232 rb_links = malloc(t->num_rbtrees * sizeof(struct rn_node**));
1235 return -E_OSL_NOMEM;
1237 if (t->num_mapped_columns) {
1238 new_row_index = malloc(t->row_index_size);
1239 if (!new_row_index) {
1242 return -E_OSL_NOMEM;
1245 /* pass 1: sanity checks */
1246 // DEBUG_LOG("sanity tests: %p:%p\n", objects[0].data,
1247 // objects[1].data);
1248 FOR_EACH_COLUMN(i, t->desc, cd) {
1249 enum osl_storage_type st = cd->storage_type;
1250 enum osl_storage_flags sf = cd->storage_flags;
1252 // ret = -E_OSL_NULL_OBJECT;
1255 if (st == OSL_DISK_STORAGE)
1257 if (sf & OSL_RBTREE) {
1258 unsigned rbtree_num = t->columns[i].rbtree_num;
1259 ret = -E_OSL_RB_KEY_EXISTS;
1260 // DEBUG_LOG("checking whether %p exists\n",
1261 // objects[i].data);
1262 if (search_rbtree(objects + i, t, i,
1263 &rb_parents[rbtree_num],
1264 &rb_links[rbtree_num]) > 0)
1267 if (sf & OSL_FIXED_SIZE) {
1268 // DEBUG_LOG("fixed size. need: %zu, have: %d\n",
1269 // objects[i].size, cd->data_size);
1270 ret = -E_OSL_BAD_DATA_SIZE;
1271 if (objects[i].size != cd->data_size)
1275 if (t->num_disk_storage_columns) {
1276 ds_name = disk_storage_name_of_object(t,
1277 &objects[t->disk_storage_name_column]);
1282 ret = unmap_table(t, OSL_MARK_CLEAN);
1285 // DEBUG_LOG("sanity tests passed%s\n", "");
1286 /* pass 2: create data files, append map data */
1287 FOR_EACH_COLUMN(i, t->desc, cd) {
1288 enum osl_storage_type st = cd->storage_type;
1289 if (st == OSL_NO_STORAGE)
1291 if (st == OSL_MAPPED_STORAGE) {
1293 struct osl_column *col = &t->columns[i];
1294 // DEBUG_LOG("appending object of size %zu\n",
1295 // objects[i].size);
1296 ret = append_map_file(t, i, objects + i, &new_size);
1299 update_cell_index(new_row_index, col, new_size,
1304 ret = write_disk_storage_file(t, i, objects + i, ds_name);
1308 ret = append_row_index(t, new_row_index);
1311 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1312 if (ret < 0) { /* truncate index and rollback changes */
1313 char *filename = index_filename(t->desc);
1315 truncate_file(filename, t->row_index_size);
1319 /* pass 3: add entry to rbtrees */
1320 if (t->num_volatile_columns) {
1322 volatile_objs = calloc(t->num_volatile_columns,
1323 sizeof(struct osl_object));
1326 FOR_EACH_VOLATILE_COLUMN(i, t, cd)
1327 volatile_objs[t->columns[i].volatile_num] = objects[i];
1330 // DEBUG_LOG("adding new entry as row #%d\n", t->num_rows - 1);
1331 ret = add_row_to_rbtrees(t, t->num_rows - 1, volatile_objs, row);
1334 // DEBUG_LOG("added new entry as row #%d\n", t->num_rows - 1);
1337 rollback: /* rollback all changes made, ignore further errors */
1338 for (i--; i >= 0; i--) {
1339 cd = get_column_description(t->desc, i);
1340 enum osl_storage_type st = cd->storage_type;
1341 if (st == OSL_NO_STORAGE)
1344 if (st == OSL_MAPPED_STORAGE)
1345 truncate_mapped_file(t, i, objects[i].size);
1346 else /* disk storage */
1347 delete_disk_storage_file(t, i, ds_name);
1349 /* ignore error and return previous error value */
1350 map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1352 free(new_row_index);
1359 __export int osl_add_row(struct osl_table *t, struct osl_object *objects)
1361 return osl_add_and_get_row(t, objects, NULL);
1364 __export int osl_get_object(const struct osl_table *t, const struct osl_row *r,
1365 unsigned col_num, struct osl_object *object)
1367 const struct osl_column_description *cd;
1370 return -E_OSL_BAD_TABLE;
1371 cd = get_column_description(t->desc, col_num);
1372 /* col must not be disk storage */
1373 if (cd->storage_type == OSL_DISK_STORAGE)
1374 return -E_OSL_BAD_STORAGE_TYPE;
1375 if (cd->storage_type == OSL_MAPPED_STORAGE)
1376 return get_mapped_object(t, col_num, r->num, object);
1378 *object = r->volatile_objects[t->columns[col_num].volatile_num];
1382 __export int osl_del_row(struct osl_table *t, struct osl_row *row)
1384 struct osl_row *r = row;
1386 const struct osl_column_description *cd;
1389 return -E_OSL_BAD_TABLE;
1390 INFO_LOG("deleting row %p\n", row);
1392 if (t->num_disk_storage_columns) {
1394 ret = disk_storage_name_of_row(t, r, &ds_name);
1397 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd)
1398 delete_disk_storage_file(t, i, ds_name);
1401 FOR_EACH_COLUMN(i, t->desc, cd) {
1402 struct osl_column *col = t->columns + i;
1403 enum osl_storage_type st = cd->storage_type;
1404 remove_rb_node(t, i, r);
1405 if (st == OSL_MAPPED_STORAGE)
1407 if (st == OSL_NO_STORAGE && !(cd->storage_flags & OSL_DONT_FREE))
1408 free(r->volatile_objects[col->volatile_num].data);
1410 if (t->num_mapped_columns) {
1411 ret = mark_row_invalid(t, r->num);
1414 t->num_invalid_rows++;
1419 free(r->volatile_objects);
1424 /* test if column has an rbtree */
1425 static int check_rbtree_col(const struct osl_table *t, unsigned col_num,
1426 struct osl_column **col)
1429 return -E_OSL_BAD_TABLE;
1430 if (!(get_column_description(t->desc, col_num)->storage_flags & OSL_RBTREE))
1431 return -E_OSL_BAD_STORAGE_FLAGS;
1432 *col = t->columns + col_num;
1436 __export int osl_get_row(const struct osl_table *t, unsigned col_num,
1437 const struct osl_object *obj, struct osl_row **result)
1440 struct rb_node *node;
1441 struct osl_row *row;
1442 struct osl_column *col;
1445 ret = check_rbtree_col(t, col_num, &col);
1448 ret = search_rbtree(obj, t, col_num, &node, NULL);
1451 row = get_row_pointer(node, t->columns[col_num].rbtree_num);
1456 static int rbtree_loop(struct osl_column *col, void *private_data,
1457 osl_rbtree_loop_func *func)
1459 struct rb_node *n, *tmp;
1461 /* this for-loop is safe against removal of an entry */
1462 for (n = rb_first(&col->rbtree), tmp = n? rb_next(n) : NULL;
1464 n = tmp, tmp = tmp? rb_next(tmp) : NULL) {
1465 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1466 if (func(r, private_data) < 0)
1472 static int rbtree_loop_reverse(struct osl_column *col, void *private_data,
1473 osl_rbtree_loop_func *func)
1475 struct rb_node *n, *tmp;
1477 /* safe against removal of an entry */
1478 for (n = rb_last(&col->rbtree), tmp = n? rb_prev(n) : NULL;
1480 n = tmp, tmp = tmp? rb_prev(tmp) : NULL) {
1481 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1482 if (func(r, private_data) < 0)
1488 __export int osl_rbtree_loop(const struct osl_table *t, unsigned col_num,
1489 void *private_data, osl_rbtree_loop_func *func)
1491 struct osl_column *col;
1493 int ret = check_rbtree_col(t, col_num, &col);
1496 return rbtree_loop(col, private_data, func);
1499 __export int osl_rbtree_loop_reverse(const struct osl_table *t, unsigned col_num,
1500 void *private_data, osl_rbtree_loop_func *func)
1502 struct osl_column *col;
1504 int ret = check_rbtree_col(t, col_num, &col);
1507 return rbtree_loop_reverse(col, private_data, func);
1510 /* TODO: Rollback changes on errors */
1511 static int rename_disk_storage_objects(struct osl_table *t,
1512 struct osl_object *old_obj, struct osl_object *new_obj)
1515 const struct osl_column_description *cd;
1516 char *old_ds_name, *new_ds_name;
1518 if (!t->num_disk_storage_columns)
1519 return 1; /* nothing to do */
1520 if (old_obj->size == new_obj->size && !memcmp(new_obj->data,
1521 old_obj->data, new_obj->size))
1522 return 1; /* object did not change */
1523 old_ds_name = disk_storage_name_of_object(t, old_obj);
1524 new_ds_name = disk_storage_name_of_object(t, new_obj);
1526 if (!old_ds_name || ! new_ds_name)
1529 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1530 char *old_filename, *new_filename;
1531 ret = create_disk_storage_object_dir(t, i, new_ds_name);
1534 old_filename = disk_storage_path(t, i, old_ds_name);
1535 new_filename = disk_storage_path(t, i, new_ds_name);
1536 if (!old_filename || !new_filename)
1539 ret = osl_rename(old_filename, new_filename);
1553 __export int osl_update_object(struct osl_table *t, const struct osl_row *r,
1554 unsigned col_num, struct osl_object *obj)
1556 struct osl_column *col;
1557 const struct osl_column_description *cd;
1561 return -E_OSL_BAD_TABLE;
1562 col = &t->columns[col_num];
1563 cd = get_column_description(t->desc, col_num);
1564 DEBUG_LOG("updating column %u of %s\n", col_num, t->desc->name);
1565 if (cd->storage_flags & OSL_RBTREE) {
1566 if (search_rbtree(obj, t, col_num, NULL, NULL) > 0)
1567 return -E_OSL_RB_KEY_EXISTS;
1569 if (cd->storage_flags & OSL_FIXED_SIZE) {
1570 if (obj->size != cd->data_size)
1571 return -E_OSL_BAD_DATA_SIZE;
1573 remove_rb_node(t, col_num, r);
1574 if (cd->storage_type == OSL_NO_STORAGE) { /* TODO: If fixed size, reuse object? */
1575 free(r->volatile_objects[col->volatile_num].data);
1576 r->volatile_objects[col->volatile_num] = *obj;
1577 } else if (cd->storage_type == OSL_DISK_STORAGE) {
1579 ret = disk_storage_name_of_row(t, r, &ds_name);
1582 ret = delete_disk_storage_file(t, col_num, ds_name);
1583 if (ret < 0 && ret != -E_OSL_NOENT) {
1587 ret = write_disk_storage_file(t, col_num, obj, ds_name);
1591 } else { /* mapped storage */
1592 struct osl_object old_obj;
1593 ret = get_mapped_object(t, col_num, r->num, &old_obj);
1597 * If the updated column is the disk storage name column, the
1598 * disk storage name changes, so we have to rename all disk
1599 * storage objects accordingly.
1601 if (col_num == t->disk_storage_name_column) {
1602 ret = rename_disk_storage_objects(t, &old_obj, obj);
1606 if (cd->storage_flags & OSL_FIXED_SIZE)
1607 memcpy(old_obj.data, obj->data, cd->data_size);
1608 else { /* TODO: if the size doesn't change, use old space */
1609 uint32_t new_data_map_size;
1611 ret = get_row_index(t, r->num, &row_index);
1614 unmap_column(t, col_num);
1615 ret = append_map_file(t, col_num, obj,
1616 &new_data_map_size);
1619 ret = map_column(t, col_num);
1622 update_cell_index(row_index, col, new_data_map_size,
1626 if (cd->storage_flags & OSL_RBTREE) {
1627 ret = insert_rbtree(t, col_num, r, obj);
1634 __export int osl_open_disk_object(const struct osl_table *t, const struct osl_row *r,
1635 unsigned col_num, struct osl_object *obj)
1637 const struct osl_column_description *cd;
1638 char *ds_name, *filename;
1642 return -E_OSL_BAD_TABLE;
1643 cd = get_column_description(t->desc, col_num);
1644 if (cd->storage_type != OSL_DISK_STORAGE)
1645 return -E_OSL_BAD_STORAGE_TYPE;
1647 ret = disk_storage_name_of_row(t, r, &ds_name);
1650 filename = disk_storage_path(t, col_num, ds_name);
1653 return -E_OSL_NOMEM;
1654 DEBUG_LOG("filename: %s\n", filename);
1655 ret = mmap_full_file(filename, O_RDONLY, &obj->data, &obj->size, NULL);
1660 __export int osl_close_disk_object(struct osl_object *obj)
1662 return osl_munmap(obj->data, obj->size);
1665 __export int osl_get_num_rows(const struct osl_table *t, unsigned *num_rows)
1668 return -E_OSL_BAD_TABLE;
1669 assert(t->num_rows >= t->num_invalid_rows);
1670 *num_rows = t->num_rows - t->num_invalid_rows;
1674 __export int osl_get_rank(const struct osl_table *t, struct osl_row *r,
1675 unsigned col_num, unsigned *rank)
1677 struct osl_object obj;
1678 struct osl_column *col;
1679 struct rb_node *node;
1680 int ret = check_rbtree_col(t, col_num, &col);
1684 ret = osl_get_object(t, r, col_num, &obj);
1687 ret = search_rbtree(&obj, t, col_num, &node, NULL);
1690 ret = rb_rank(node, rank);
1692 return -E_OSL_BAD_ROW;
1696 __export int osl_get_nth_row(const struct osl_table *t, unsigned col_num,
1697 unsigned n, struct osl_row **result)
1699 struct osl_column *col;
1700 struct rb_node *node;
1705 return -E_OSL_RB_KEY_NOT_FOUND;
1706 ret = osl_get_num_rows(t, &num_rows);
1710 return -E_OSL_RB_KEY_NOT_FOUND;
1711 ret = check_rbtree_col(t, col_num, &col);
1714 node = rb_nth(col->rbtree.rb_node, n);
1716 return -E_OSL_RB_KEY_NOT_FOUND;
1717 *result = get_row_pointer(node, col->rbtree_num);
1721 __export int osl_rbtree_first_row(const struct osl_table *t, unsigned col_num,
1722 struct osl_row **result)
1724 return osl_get_nth_row(t, col_num, 1, result);
1727 __export int osl_rbtree_last_row(const struct osl_table *t, unsigned col_num,
1728 struct osl_row **result)
1731 int ret = osl_get_num_rows(t, &num_rows);
1735 return osl_get_nth_row(t, col_num, num_rows, result);