2 * Copyright (C) 2007-2008 Andre Noll <maan@systemlinux.org>
4 * Licensed under the GPL v2. For licencing details see COPYING.
7 /** \file osl.c Object storage layer functions. */
8 #include <dirent.h> /* readdir() */
17 /* Taken from Drepper: How to write shared libraries, Appendix B. */
19 #define MSGSTRFIELD(line) MSGSTRFIELD1(line)
20 #define MSGSTRFIELD1(line) str##line
21 static const union msgstr_t {
23 #define _S(n, s) char MSGSTRFIELD(__LINE__)[sizeof(s)];
33 static const unsigned int errmsgidx[] = {
34 #define _S(n, s) [n] = offsetof(union msgstr_t, MSGSTRFIELD(__LINE__)),
39 __export const char *osl_strerror(int num)
41 if (IS_SYSTEM_ERROR(num))
42 return strerror((num) & ((1 << SYSTEM_ERROR_BIT) - 1));
43 return msgstr.str + errmsgidx[num];
48 static void __attribute ((constructor)) init_loglevel(void)
50 char *p = getenv("OSL_LOGLEVEL");
52 /* don't log anything if unset */
53 loglevel = p? atoi(p) : EMERG + 1;
60 * \param fmt Usual format string.
62 * All XXX_LOG() macros use this function.
64 __printf_2_3 void __log(int ll, const char* fmt,...)
77 strftime(str, sizeof(str), "%b %d %H:%M:%S", tm);
78 fprintf(outfd, "%s ", str);
80 vfprintf(outfd, fmt, argp);
85 * A wrapper for lseek(2).
87 * \param fd The file descriptor whose offset is to be to repositioned.
88 * \param offset A value-result parameter.
89 * \param whence Usual repositioning directive.
91 * Reposition the offset of the file descriptor \a fd to the argument \a offset
92 * according to the directive \a whence. Upon successful return, \a offset
93 * contains the resulting offset location as measured in bytes from the
94 * beginning of the file.
96 * \return Positive on success. Otherwise, the function returns \p -E_OSL_LSEEK.
100 static int __lseek(int fd, off_t *offset, int whence)
102 *offset = lseek(fd, *offset, whence);
103 int ret = -E_OSL_LSEEK;
109 static int append_file(const char *filename, char *header, size_t header_size,
110 char *data, size_t data_size, uint32_t *new_pos)
114 // DEBUG_LOG("appending %zu + %zu bytes\n", header_size, data_size);
115 ret = osl_open(filename, O_WRONLY | O_CREAT | O_APPEND, 0644);
119 if (header && header_size) {
120 ret = write_all(fd, header, &header_size);
124 ret = write_all(fd, data, &data_size);
129 ret = __lseek(fd, &offset, SEEK_END);
132 // DEBUG_LOG("new file size: " FMT_OFF_T "\n", offset);
141 static int verify_name(const char *name)
144 return -E_OSL_BAD_NAME;
146 return -E_OSL_BAD_NAME;
147 if (strchr(name, '/'))
148 return -E_OSL_BAD_NAME;
149 if (!strcmp(name, ".."))
150 return -E_OSL_BAD_NAME;
151 if (!strcmp(name, "."))
152 return -E_OSL_BAD_NAME;
156 int osl_hash_compare(const struct osl_object *obj1, const struct osl_object *obj2)
158 return hash_compare((HASH_TYPE *)obj1->data, (HASH_TYPE *)obj2->data);
161 static char *disk_storage_dirname(const struct osl_table *t, unsigned col_num,
164 char *dirname, *column_name = column_filename(t, col_num);
168 if (!(t->desc->flags & OSL_LARGE_TABLE))
170 dirname = make_message("%s/%.2s", column_name, ds_name);
175 static char *disk_storage_name_of_object(const struct osl_table *t,
176 const struct osl_object *obj)
178 HASH_TYPE hash[HASH_SIZE];
179 hash_object(obj, hash);
180 return disk_storage_name_of_hash(t, hash);
183 static int disk_storage_name_of_row(const struct osl_table *t,
184 const struct osl_row *row, char **name)
186 struct osl_object obj;
187 int ret = osl_get_object(t, row, t->disk_storage_name_column, &obj);
191 *name = disk_storage_name_of_object(t, &obj);
194 return -ERRNO_TO_ERROR(ENOMEM);
197 static void column_name_hash(const char *col_name, HASH_TYPE *hash)
199 hash_function(col_name, strlen(col_name), hash);
202 static int init_column_descriptions(struct osl_table *t)
205 const struct osl_column_description *cd;
207 ret = -E_OSL_BAD_TABLE_DESC;
208 ret = verify_name(t->desc->name);
211 ret = -E_OSL_BAD_DB_DIR;
212 if (!t->desc->dir && (t->num_disk_storage_columns || t->num_mapped_columns))
214 /* the size of the index header without column descriptions */
215 t->index_header_size = IDX_COLUMN_DESCRIPTIONS;
216 FOR_EACH_COLUMN(i, t->desc, cd) {
217 struct osl_column *col = t->columns + i;
218 if (cd->storage_flags & OSL_RBTREE) {
219 if (!cd->compare_function)
220 return -E_OSL_NO_COMPARE_FUNC;
222 if (cd->storage_type == OSL_NO_STORAGE)
224 ret = -E_OSL_NO_COLUMN_NAME;
225 if (!cd->name || !cd->name[0])
227 ret = verify_name(cd->name);
230 t->index_header_size += index_column_description_size(cd->name);
231 column_name_hash(cd->name, col->name_hash);
232 ret = -E_OSL_DUPLICATE_COL_NAME;
233 for (j = i + 1; j < t->desc->num_columns; j++) {
234 const char *name2 = get_column_description(t->desc,
236 if (cd->name && name2 && !strcmp(cd->name, name2))
246 * Initialize a struct table from given table description.
248 * \param desc The description of the osl table.
249 * \param table_ptr Result is returned here.
251 * This function performs several sanity checks on \p desc and returns if any
252 * of these tests fail. On success, a struct \p osl_table is allocated and
253 * initialized with data derived from \p desc.
257 * \sa struct osl_table.
259 int init_table_structure(const struct osl_table_description *desc,
260 struct osl_table **table_ptr)
262 const struct osl_column_description *cd;
263 struct osl_table *t = calloc(1, sizeof(*t));
264 int i, ret = -ERRNO_TO_ERROR(ENOMEM), have_disk_storage_name_column = 0;
268 ret = -E_OSL_BAD_TABLE_DESC;
271 DEBUG_LOG("creating table structure for '%s' from table "
272 "description\n", desc->name);
273 ret = -E_OSL_NO_COLUMN_DESC;
274 if (!desc->column_descriptions)
276 ret = -E_OSL_NO_COLUMNS;
277 if (!desc->num_columns)
279 ret = -ERRNO_TO_ERROR(ENOMEM);
280 t->columns = calloc(desc->num_columns, sizeof(struct osl_column));
284 FOR_EACH_COLUMN(i, t->desc, cd) {
285 enum osl_storage_type st = cd->storage_type;
286 enum osl_storage_flags sf = cd->storage_flags;
287 struct osl_column *col = &t->columns[i];
289 ret = -E_OSL_BAD_STORAGE_TYPE;
290 if (st != OSL_MAPPED_STORAGE && st != OSL_DISK_STORAGE
291 && st != OSL_NO_STORAGE)
293 ret = -E_OSL_BAD_STORAGE_FLAGS;
294 if (st == OSL_DISK_STORAGE && sf & OSL_RBTREE)
296 ret = -E_OSL_BAD_STORAGE_SIZE;
297 if (sf & OSL_FIXED_SIZE && !cd->data_size)
300 case OSL_DISK_STORAGE:
301 t->num_disk_storage_columns++;
303 case OSL_MAPPED_STORAGE:
304 t->num_mapped_columns++;
305 col->index_offset = t->row_index_size;
306 t->row_index_size += 8;
309 col->volatile_num = t->num_volatile_columns;
310 t->num_volatile_columns++;
313 if (sf & OSL_RBTREE) {
314 col->rbtree_num = t->num_rbtrees;
316 if ((sf & OSL_UNIQUE) && (st == OSL_MAPPED_STORAGE)) {
317 if (!have_disk_storage_name_column)
318 t->disk_storage_name_column = i;
319 have_disk_storage_name_column = 1;
323 ret = -E_OSL_NO_UNIQUE_RBTREE_COLUMN;
324 if (t->num_disk_storage_columns && !have_disk_storage_name_column)
326 ret = -E_OSL_NO_RBTREE_COL;
330 DEBUG_LOG("OK. Index entry size: %u\n", t->row_index_size);
331 ret = init_column_descriptions(t);
343 * Read the table description from index header.
345 * \param map The memory mapping of the index file.
346 * \param desc The values found in the index header are returned here.
348 * Read the index header, check for the paraslash magic string and the table version number.
349 * Read all information stored in the index header into \a desc.
353 * \sa struct osl_table_description, osl_create_table.
355 int read_table_desc(struct osl_object *map, struct osl_table_description *desc)
357 char *buf = map->data;
359 uint16_t header_size;
362 struct osl_column_description *cd;
364 if (map->size < MIN_INDEX_HEADER_SIZE(1))
365 return -E_OSL_SHORT_TABLE;
366 if (strncmp(buf + IDX_PARA_MAGIC, PARA_MAGIC, strlen(PARA_MAGIC)))
367 return -E_OSL_NO_MAGIC;
368 version = read_u8(buf + IDX_VERSION);
369 if (version < MIN_TABLE_VERSION || version > MAX_TABLE_VERSION)
370 return -E_OSL_VERSION_MISMATCH;
371 desc->flags = read_u8(buf + IDX_TABLE_FLAGS);
372 desc->num_columns = read_u16(buf + IDX_NUM_COLUMNS);
373 INFO_LOG("%u columns\n", desc->num_columns);
374 if (!desc->num_columns)
375 return -E_OSL_NO_COLUMNS;
376 header_size = read_u16(buf + IDX_HEADER_SIZE);
377 if (map->size < header_size)
378 return -E_OSL_BAD_SIZE;
379 desc->column_descriptions = calloc(desc->num_columns,
380 sizeof(struct osl_column_description));
381 if (!desc->column_descriptions)
382 return -ERRNO_TO_ERROR(ENOMEM);
383 offset = IDX_COLUMN_DESCRIPTIONS;
384 FOR_EACH_COLUMN(i, desc, cd) {
387 ret = -E_OSL_SHORT_TABLE;
388 if (map->size < offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE) {
389 ERROR_LOG("map size = %zu < %u = offset + min desc size\n",
390 map->size, offset + MIN_IDX_COLUMN_DESCRIPTION_SIZE);
393 cd->storage_type = read_u16(buf + offset + IDX_CD_STORAGE_TYPE);
394 cd->storage_flags = read_u16(buf + offset +
395 IDX_CD_STORAGE_FLAGS);
396 cd->data_size = read_u32(buf + offset + IDX_CD_DATA_SIZE);
397 null_byte = memchr(buf + offset + IDX_CD_NAME, '\0',
398 map->size - offset - IDX_CD_NAME);
399 ret = -E_OSL_INDEX_CORRUPTION;
402 ret = -ERRNO_TO_ERROR(ENOMEM);
403 cd->name = strdup(buf + offset + IDX_CD_NAME);
406 offset += index_column_description_size(cd->name);
408 if (offset != header_size) {
409 ret = -E_OSL_INDEX_CORRUPTION;
410 ERROR_LOG("real header size = %u != %u = stored header size\n",
411 offset, header_size);
416 FOR_EACH_COLUMN(i, desc, cd)
422 * check whether the table description given by \p t->desc matches the on-disk
423 * table structure stored in the index of \a t.
425 static int compare_table_descriptions(struct osl_table *t)
428 struct osl_table_description desc;
429 const struct osl_column_description *cd1, *cd2;
431 /* read the on-disk structure into desc */
432 ret = read_table_desc(&t->index_map, &desc);
435 ret = -E_OSL_BAD_TABLE_FLAGS;
436 if (desc.flags != t->desc->flags)
438 ret = -E_OSL_BAD_COLUMN_NUM;
439 if (desc.num_columns > t->desc->num_columns)
441 if (desc.num_columns < t->desc->num_columns) {
442 struct osl_column_description *cd;
443 unsigned diff = t->desc->num_columns - desc.num_columns;
444 INFO_LOG("extending table by %u volatile columns\n", diff);
445 ret = -ERRNO_TO_ERROR(ENOMEM);
446 desc.column_descriptions = realloc(desc.column_descriptions,
447 t->desc->num_columns * sizeof(struct osl_column_description));
448 if (!desc.column_descriptions)
450 for (i = desc.num_columns; i < t->desc->num_columns; i++) {
451 cd = get_column_description(&desc, i);
452 cd->storage_type = OSL_NO_STORAGE;
455 desc.num_columns += diff;
457 FOR_EACH_COLUMN(i, t->desc, cd1) {
458 cd2 = get_column_description(&desc, i);
459 ret = -E_OSL_BAD_STORAGE_TYPE;
460 if (cd1->storage_type != cd2->storage_type)
462 if (cd1->storage_type == OSL_NO_STORAGE)
464 ret = -E_OSL_BAD_STORAGE_FLAGS;
465 if (cd1->storage_flags != cd2->storage_flags) {
466 ERROR_LOG("sf1 = %u != %u = sf2\n",
467 cd1->storage_flags, cd2->storage_flags);
470 ret = -E_OSL_BAD_DATA_SIZE;
471 if (cd1->storage_flags & OSL_FIXED_SIZE)
472 if (cd1->data_size != cd2->data_size)
474 ret = -E_OSL_BAD_COLUMN_NAME;
475 if (strcmp(cd1->name, cd2->name))
478 INFO_LOG("table description of '%s' matches on-disk data, good\n",
482 FOR_EACH_COLUMN(i, &desc, cd1)
484 free(desc.column_descriptions);
488 static int create_table_index(struct osl_table *t)
490 char *buf, *filename;
492 size_t size = t->index_header_size;
493 const struct osl_column_description *cd;
496 INFO_LOG("creating %zu byte index for table %s\n", size,
498 buf = calloc(1, size);
500 return -ERRNO_TO_ERROR(ENOMEM);
501 sprintf(buf + IDX_PARA_MAGIC, "%s", PARA_MAGIC);
502 write_u8(buf + IDX_TABLE_FLAGS, t->desc->flags);
503 write_u8(buf + IDX_DIRTY_FLAG, 0);
504 write_u8(buf + IDX_VERSION, CURRENT_TABLE_VERSION);
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);
527 ret = -ERRNO_TO_ERROR(ENOMEM);
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 && !is_errno(-ret, EEXIST))
550 table_dir = make_message("%s/%s", desc->dir,
552 ret = -ERRNO_TO_ERROR(ENOMEM);
555 ret = osl_mkdir(table_dir, 0777);
559 ret = -ERRNO_TO_ERROR(ENOMEM);
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);
663 return -ERRNO_TO_ERROR(ENOMEM);
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);
704 return -ERRNO_TO_ERROR(ENOMEM);
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];
769 if (t->num_rows <= row_num)
770 return -E_OSL_BAD_ROW_NUM;
771 ret = get_cell_index(t, row_num, col_num, &cell_index);
774 offset = read_u32(cell_index);
775 obj->size = read_u32(cell_index + 4) - 1;
776 header = col->data_map.data + offset;
777 obj->data = header + 1;
778 if (read_u8(header) == 0xff) {
779 ERROR_LOG("col %u, size %zu, offset %u\n", col_num,
781 return -E_OSL_INVALID_OBJECT;
786 static int search_rbtree(const struct osl_object *obj,
787 const struct osl_table *t, unsigned col_num,
788 struct rb_node **result, struct rb_node ***rb_link)
790 struct osl_column *col = &t->columns[col_num];
791 struct rb_node **new = &col->rbtree.rb_node, *parent = NULL;
792 const struct osl_column_description *cd =
793 get_column_description(t->desc, col_num);
794 enum osl_storage_type st = cd->storage_type;
796 struct osl_row *this_row = get_row_pointer(*new,
799 struct osl_object this_obj;
801 if (st == OSL_MAPPED_STORAGE) {
802 ret = get_mapped_object(t, col_num, this_row->num,
807 this_obj = this_row->volatile_objects[col->volatile_num];
808 ret = cd->compare_function(obj, &this_obj);
811 *result = get_rb_node_pointer(this_row,
816 new = &((*new)->rb_left);
818 new = &((*new)->rb_right);
824 return -E_OSL_RB_KEY_NOT_FOUND;
827 static int insert_rbtree(struct osl_table *t, unsigned col_num,
828 const struct osl_row *row, const struct osl_object *obj)
830 struct rb_node *parent, **rb_link;
833 int ret = search_rbtree(obj, t, col_num, &parent, &rb_link);
836 return -E_OSL_RB_KEY_EXISTS;
837 rbtree_num = t->columns[col_num].rbtree_num;
838 n = get_rb_node_pointer(row, rbtree_num);
839 rb_link_node(n, parent, rb_link);
840 rb_insert_color(n, &t->columns[col_num].rbtree);
844 static void remove_rb_node(struct osl_table *t, unsigned col_num,
845 const struct osl_row *row)
847 struct osl_column *col = &t->columns[col_num];
848 const struct osl_column_description *cd =
849 get_column_description(t->desc, col_num);
850 enum osl_storage_flags sf = cd->storage_flags;
851 struct rb_node *victim, *splice_out_node, *tmp;
852 if (!(sf & OSL_RBTREE))
855 * Which node is removed/spliced out actually depends on how many
856 * children the victim node has: If it has no children, it gets
857 * deleted. If it has one child, it gets spliced out. If it has two
858 * children, its successor (which has at most a right child) gets
861 victim = get_rb_node_pointer(row, col->rbtree_num);
862 if (victim->rb_left && victim->rb_right)
863 splice_out_node = rb_next(victim);
865 splice_out_node = victim;
866 /* Go up to the root and decrement the size of each node in the path. */
867 for (tmp = splice_out_node; tmp; tmp = rb_parent(tmp))
869 rb_erase(victim, &col->rbtree);
872 static int add_row_to_rbtrees(struct osl_table *t, uint32_t row_num,
873 struct osl_object *volatile_objs, struct osl_row **row_ptr)
877 struct osl_row *row = allocate_row(t->num_rbtrees);
878 const struct osl_column_description *cd;
881 return -ERRNO_TO_ERROR(ENOMEM);
883 row->volatile_objects = volatile_objs;
884 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
885 if (cd->storage_type == OSL_MAPPED_STORAGE) {
886 struct osl_object obj;
887 ret = get_mapped_object(t, i, row_num, &obj);
890 ret = insert_rbtree(t, i, row, &obj);
891 } else { /* volatile */
892 const struct osl_object *obj
893 = volatile_objs + t->columns[i].volatile_num;
894 ret = insert_rbtree(t, i, row, obj);
902 err: /* rollback changes, i.e. remove added entries from rbtrees */
904 remove_rb_node(t, i--, row);
909 static void free_volatile_objects(const struct osl_table *t,
910 enum osl_close_flags flags)
914 struct osl_column *rb_col;
915 const struct osl_column_description *cd;
917 if (!t->num_volatile_columns)
919 /* find the first rbtree column (any will do) */
920 FOR_EACH_RBTREE_COLUMN(i, t, cd)
922 rb_col = t->columns + i;
923 /* walk that rbtree and free all volatile objects */
924 for (n = rb_first(&rb_col->rbtree); n; n = rb_next(n)) {
925 struct osl_row *r = get_row_pointer(n, rb_col->rbtree_num);
926 if (flags & OSL_FREE_VOLATILE)
927 FOR_EACH_VOLATILE_COLUMN(j, t, cd) {
928 if (cd->storage_flags & OSL_DONT_FREE)
930 free(r->volatile_objects[
931 t->columns[j].volatile_num].data);
933 // for (j = 0; j < t->num_volatile_columns; j++)
934 // free(r->volatile_objects[j].data);
935 free(r->volatile_objects);
940 * Erase all rbtree nodes and free resources.
942 * \param t Pointer to an open osl table.
944 * This function is called by osl_close_table().
946 void clear_rbtrees(struct osl_table *t)
948 const struct osl_column_description *cd;
949 unsigned i, rbtrees_cleared = 0;
951 FOR_EACH_RBTREE_COLUMN(i, t, cd) {
952 struct osl_column *col = &t->columns[i];
955 for (n = rb_first(&col->rbtree); n;) {
957 rb_erase(n, &col->rbtree);
958 if (rbtrees_cleared == t->num_rbtrees) {
959 r = get_row_pointer(n, col->rbtree_num);
969 __export int osl_close_table(struct osl_table *t, enum osl_close_flags flags)
974 return -E_OSL_BAD_TABLE;
975 NOTICE_LOG("closing table %s\n", t->desc->name);
976 free_volatile_objects(t, flags);
978 ret = unmap_table(t, flags);
980 ERROR_LOG("unmap_table failed: %d\n", ret);
987 * Find out whether the given row number corresponds to an invalid row.
989 * \param t Pointer to the osl table.
990 * \param row_num The number of the row in question.
992 * By definition, a row is considered invalid if all its index entries
995 * \return Positive if \a row_num corresponds to an invalid row,
996 * zero if it corresponds to a valid row, negative on errors.
998 int row_is_invalid(struct osl_table *t, uint32_t row_num)
1001 int i, ret = get_row_index(t, row_num, &row_index);
1005 for (i = 0; i < t->row_index_size; i++) {
1006 if ((unsigned char)row_index[i] != 0xff)
1009 INFO_LOG("row %d is invalid\n", row_num);
1014 * Invalidate a row of an osl table.
1016 * \param t Pointer to an open osl table.
1017 * \param row_num Number of the row to mark as invalid.
1019 * This function marks each mapped object in the index entry of \a row as
1024 int mark_row_invalid(struct osl_table *t, uint32_t row_num)
1027 int ret = get_row_index(t, row_num, &row_index);
1031 INFO_LOG("marking row %d as invalid\n", row_num);
1032 memset(row_index, 0xff, t->row_index_size);
1037 * Initialize all rbtrees and compute number of invalid rows.
1039 * \param t The table containing the rbtrees to be initialized.
1043 int init_rbtrees(struct osl_table *t)
1046 const struct osl_column_description *cd;
1048 /* create rbtrees */
1049 FOR_EACH_RBTREE_COLUMN(i, t, cd)
1050 t->columns[i].rbtree = RB_ROOT;
1051 /* add valid rows to rbtrees */
1052 t->num_invalid_rows = 0;
1053 for (i = 0; i < t->num_rows; i++) {
1054 ret = row_is_invalid(t, i);
1058 t->num_invalid_rows++;
1061 ret = add_row_to_rbtrees(t, i, NULL, NULL);
1068 __export int osl_open_table(const struct osl_table_description *table_desc,
1069 struct osl_table **result)
1072 struct osl_table *t;
1073 const struct osl_column_description *cd;
1075 NOTICE_LOG("opening table %s\n", table_desc->name);
1076 ret = init_table_structure(table_desc, &t);
1079 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1080 struct stat statbuf;
1081 char *dirname = column_filename(t, i);
1083 ret = -ERRNO_TO_ERROR(ENOMEM);
1086 /* check if directory exists */
1087 ret = stat(dirname, &statbuf);
1090 ret = -ERRNO_TO_ERROR(errno);
1093 ret = -ERRNO_TO_ERROR(ENOTDIR);
1094 if (!S_ISDIR(statbuf.st_mode))
1097 ret = map_table(t, MAP_TBL_FL_VERIFY_INDEX);
1101 DEBUG_LOG("num rows: %d\n", t->num_rows);
1102 ret = init_rbtrees(t);
1104 osl_close_table(t, OSL_MARK_CLEAN); /* ignore further errors */
1115 static int create_disk_storage_object_dir(const struct osl_table *t,
1116 unsigned col_num, const char *ds_name)
1121 if (!(t->desc->flags & OSL_LARGE_TABLE))
1123 dirname = disk_storage_dirname(t, col_num, ds_name);
1125 return -ERRNO_TO_ERROR(ENOMEM);
1126 ret = osl_mkdir(dirname, 0777);
1128 if (ret < 0 && !is_errno(-ret, EEXIST))
1133 static int write_disk_storage_file(const struct osl_table *t, unsigned col_num,
1134 const struct osl_object *obj, const char *ds_name)
1139 ret = create_disk_storage_object_dir(t, col_num, ds_name);
1142 filename = disk_storage_path(t, col_num, ds_name);
1144 return -ERRNO_TO_ERROR(ENOMEM);
1145 ret = write_file(filename, obj->data, obj->size);
1150 static int append_map_file(const struct osl_table *t, unsigned col_num,
1151 const struct osl_object *obj, uint32_t *new_size)
1153 char *filename = column_filename(t, col_num);
1155 char header = 0; /* zero means valid object */
1158 return -ERRNO_TO_ERROR(ENOMEM);
1159 ret = append_file(filename, &header, 1, obj->data, obj->size,
1165 static int append_row_index(const struct osl_table *t, char *row_index)
1170 if (!t->num_mapped_columns)
1172 filename = index_filename(t->desc);
1174 return -ERRNO_TO_ERROR(ENOMEM);
1175 ret = append_file(filename, NULL, 0, row_index,
1176 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 -ERRNO_TO_ERROR(ENOMEM);
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 -ERRNO_TO_ERROR(ENOMEM);
1202 ret = unlink(filename);
1206 return -ERRNO_TO_ERROR(err);
1207 if (!(t->desc->flags & OSL_LARGE_TABLE))
1209 dirname = disk_storage_dirname(t, col_num, ds_name);
1211 return -ERRNO_TO_ERROR(ENOMEM);
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 -ERRNO_TO_ERROR(ENOMEM);
1232 rb_links = malloc(t->num_rbtrees * sizeof(struct rn_node**));
1235 return -ERRNO_TO_ERROR(ENOMEM);
1237 if (t->num_mapped_columns) {
1238 new_row_index = malloc(t->row_index_size);
1239 if (!new_row_index) {
1242 return -ERRNO_TO_ERROR(ENOMEM);
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]);
1278 ret = -ERRNO_TO_ERROR(ENOMEM);
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) {
1321 ret = -ERRNO_TO_ERROR(ENOMEM);
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 static int mark_mapped_object_invalid(const struct osl_table *t,
1383 uint32_t row_num, unsigned col_num)
1385 struct osl_object obj;
1387 int ret = get_mapped_object(t, col_num, row_num, &obj);
1397 __export int osl_del_row(struct osl_table *t, struct osl_row *row)
1399 struct osl_row *r = row;
1401 const struct osl_column_description *cd;
1404 return -E_OSL_BAD_TABLE;
1405 INFO_LOG("deleting row %p\n", row);
1407 if (t->num_disk_storage_columns) {
1409 ret = disk_storage_name_of_row(t, r, &ds_name);
1412 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd)
1413 delete_disk_storage_file(t, i, ds_name);
1416 FOR_EACH_COLUMN(i, t->desc, cd) {
1417 struct osl_column *col = t->columns + i;
1418 enum osl_storage_type st = cd->storage_type;
1419 remove_rb_node(t, i, r);
1420 if (st == OSL_MAPPED_STORAGE) {
1421 mark_mapped_object_invalid(t, r->num, i);
1424 if (st == OSL_NO_STORAGE && !(cd->storage_flags & OSL_DONT_FREE))
1425 free(r->volatile_objects[col->volatile_num].data);
1427 if (t->num_mapped_columns) {
1428 ret = mark_row_invalid(t, r->num);
1431 t->num_invalid_rows++;
1436 free(r->volatile_objects);
1441 /* test if column has an rbtree */
1442 static int check_rbtree_col(const struct osl_table *t, unsigned col_num,
1443 struct osl_column **col)
1446 return -E_OSL_BAD_TABLE;
1447 if (!(get_column_description(t->desc, col_num)->storage_flags & OSL_RBTREE))
1448 return -E_OSL_BAD_STORAGE_FLAGS;
1449 *col = t->columns + col_num;
1453 __export int osl_get_row(const struct osl_table *t, unsigned col_num,
1454 const struct osl_object *obj, struct osl_row **result)
1457 struct rb_node *node;
1458 struct osl_row *row;
1459 struct osl_column *col;
1462 ret = check_rbtree_col(t, col_num, &col);
1465 ret = search_rbtree(obj, t, col_num, &node, NULL);
1468 row = get_row_pointer(node, t->columns[col_num].rbtree_num);
1473 static int rbtree_loop(struct osl_column *col, void *private_data,
1474 osl_rbtree_loop_func *func)
1476 struct rb_node *n, *tmp;
1478 /* this for-loop is safe against removal of an entry */
1479 for (n = rb_first(&col->rbtree), tmp = n? rb_next(n) : NULL;
1481 n = tmp, tmp = tmp? rb_next(tmp) : NULL) {
1482 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1483 if (func(r, private_data) < 0)
1489 static int rbtree_loop_reverse(struct osl_column *col, void *private_data,
1490 osl_rbtree_loop_func *func)
1492 struct rb_node *n, *tmp;
1494 /* safe against removal of an entry */
1495 for (n = rb_last(&col->rbtree), tmp = n? rb_prev(n) : NULL;
1497 n = tmp, tmp = tmp? rb_prev(tmp) : NULL) {
1498 struct osl_row *r = get_row_pointer(n, col->rbtree_num);
1499 if (func(r, private_data) < 0)
1505 __export int osl_rbtree_loop(const struct osl_table *t, unsigned col_num,
1506 void *private_data, osl_rbtree_loop_func *func)
1508 struct osl_column *col;
1510 int ret = check_rbtree_col(t, col_num, &col);
1513 return rbtree_loop(col, private_data, func);
1516 __export int osl_rbtree_loop_reverse(const struct osl_table *t, unsigned col_num,
1517 void *private_data, osl_rbtree_loop_func *func)
1519 struct osl_column *col;
1521 int ret = check_rbtree_col(t, col_num, &col);
1524 return rbtree_loop_reverse(col, private_data, func);
1527 /* TODO: Rollback changes on errors */
1528 static int rename_disk_storage_objects(struct osl_table *t,
1529 struct osl_object *old_obj, struct osl_object *new_obj)
1532 const struct osl_column_description *cd;
1533 char *old_ds_name, *new_ds_name;
1535 if (!t->num_disk_storage_columns)
1536 return 1; /* nothing to do */
1537 if (old_obj->size == new_obj->size && !memcmp(new_obj->data,
1538 old_obj->data, new_obj->size))
1539 return 1; /* object did not change */
1540 old_ds_name = disk_storage_name_of_object(t, old_obj);
1541 new_ds_name = disk_storage_name_of_object(t, new_obj);
1542 ret = -ERRNO_TO_ERROR(ENOMEM);
1543 if (!old_ds_name || ! new_ds_name)
1546 FOR_EACH_DISK_STORAGE_COLUMN(i, t, cd) {
1547 char *old_filename, *new_filename;
1548 ret = create_disk_storage_object_dir(t, i, new_ds_name);
1551 old_filename = disk_storage_path(t, i, old_ds_name);
1552 new_filename = disk_storage_path(t, i, new_ds_name);
1553 if (!old_filename || !new_filename)
1554 ret = -ERRNO_TO_ERROR(ENOMEM);
1556 ret = osl_rename(old_filename, new_filename);
1570 __export int osl_update_object(struct osl_table *t, const struct osl_row *r,
1571 unsigned col_num, struct osl_object *obj)
1573 struct osl_column *col;
1574 const struct osl_column_description *cd;
1578 return -E_OSL_BAD_TABLE;
1579 col = &t->columns[col_num];
1580 cd = get_column_description(t->desc, col_num);
1581 DEBUG_LOG("updating column %u of %s\n", col_num, t->desc->name);
1582 if (cd->storage_flags & OSL_RBTREE) {
1583 if (search_rbtree(obj, t, col_num, NULL, NULL) > 0)
1584 return -E_OSL_RB_KEY_EXISTS;
1586 if (cd->storage_flags & OSL_FIXED_SIZE) {
1587 if (obj->size != cd->data_size)
1588 return -E_OSL_BAD_DATA_SIZE;
1590 remove_rb_node(t, col_num, r);
1591 if (cd->storage_type == OSL_NO_STORAGE) { /* TODO: If fixed size, reuse object? */
1592 free(r->volatile_objects[col->volatile_num].data);
1593 r->volatile_objects[col->volatile_num] = *obj;
1594 } else if (cd->storage_type == OSL_DISK_STORAGE) {
1596 ret = disk_storage_name_of_row(t, r, &ds_name);
1599 ret = delete_disk_storage_file(t, col_num, ds_name);
1600 if (ret < 0 && !is_errno(-ret, ENOENT)) {
1604 ret = write_disk_storage_file(t, col_num, obj, ds_name);
1608 } else { /* mapped storage */
1609 struct osl_object old_obj;
1610 ret = get_mapped_object(t, col_num, r->num, &old_obj);
1614 * If the updated column is the disk storage name column, the
1615 * disk storage name changes, so we have to rename all disk
1616 * storage objects accordingly.
1618 if (col_num == t->disk_storage_name_column) {
1619 ret = rename_disk_storage_objects(t, &old_obj, obj);
1623 if (cd->storage_flags & OSL_FIXED_SIZE)
1624 memcpy(old_obj.data, obj->data, cd->data_size);
1625 else { /* TODO: if the size doesn't change, use old space */
1626 uint32_t new_data_map_size;
1628 ret = get_row_index(t, r->num, &row_index);
1631 ret = mark_mapped_object_invalid(t, r->num, col_num);
1634 unmap_column(t, col_num);
1635 ret = append_map_file(t, col_num, obj,
1636 &new_data_map_size);
1639 ret = map_column(t, col_num);
1642 update_cell_index(row_index, col, new_data_map_size,
1646 if (cd->storage_flags & OSL_RBTREE) {
1647 ret = insert_rbtree(t, col_num, r, obj);
1654 __export int osl_open_disk_object(const struct osl_table *t, const struct osl_row *r,
1655 unsigned col_num, struct osl_object *obj)
1657 const struct osl_column_description *cd;
1658 char *ds_name, *filename;
1662 return -E_OSL_BAD_TABLE;
1663 cd = get_column_description(t->desc, col_num);
1664 if (cd->storage_type != OSL_DISK_STORAGE)
1665 return -E_OSL_BAD_STORAGE_TYPE;
1667 ret = disk_storage_name_of_row(t, r, &ds_name);
1670 filename = disk_storage_path(t, col_num, ds_name);
1673 return -ERRNO_TO_ERROR(ENOMEM);
1674 DEBUG_LOG("filename: %s\n", filename);
1675 ret = mmap_full_file(filename, O_RDONLY, &obj->data, &obj->size, NULL);
1680 __export int osl_close_disk_object(struct osl_object *obj)
1682 return osl_munmap(obj->data, obj->size);
1685 __export int osl_get_num_rows(const struct osl_table *t, unsigned *num_rows)
1688 return -E_OSL_BAD_TABLE;
1689 assert(t->num_rows >= t->num_invalid_rows);
1690 *num_rows = t->num_rows - t->num_invalid_rows;
1694 __export int osl_get_rank(const struct osl_table *t, struct osl_row *r,
1695 unsigned col_num, unsigned *rank)
1697 struct osl_object obj;
1698 struct osl_column *col;
1699 struct rb_node *node;
1700 int ret = check_rbtree_col(t, col_num, &col);
1704 ret = osl_get_object(t, r, col_num, &obj);
1707 ret = search_rbtree(&obj, t, col_num, &node, NULL);
1710 ret = rb_rank(node, rank);
1712 return -E_OSL_BAD_ROW;
1716 __export int osl_get_nth_row(const struct osl_table *t, unsigned col_num,
1717 unsigned n, struct osl_row **result)
1719 struct osl_column *col;
1720 struct rb_node *node;
1725 return -E_OSL_RB_KEY_NOT_FOUND;
1726 ret = osl_get_num_rows(t, &num_rows);
1730 return -E_OSL_RB_KEY_NOT_FOUND;
1731 ret = check_rbtree_col(t, col_num, &col);
1734 node = rb_nth(col->rbtree.rb_node, n);
1736 return -E_OSL_RB_KEY_NOT_FOUND;
1737 *result = get_row_pointer(node, col->rbtree_num);
1741 __export int osl_rbtree_first_row(const struct osl_table *t, unsigned col_num,
1742 struct osl_row **result)
1744 return osl_get_nth_row(t, col_num, 1, result);
1747 __export int osl_rbtree_last_row(const struct osl_table *t, unsigned col_num,
1748 struct osl_row **result)
1751 int ret = osl_get_num_rows(t, &num_rows);
1755 return osl_get_nth_row(t, col_num, num_rows, result);